Content uploaded by Marius Thériault
Author content
All content in this area was uploaded by Marius Thériault
Content may be subject to copyright.
JRER
兩
Vol. 23
兩
Nos. 1/2 – 2002
Landscaping and House Values:
An Empirical Investigation
Authors Franc¸ois Des Rosiers, Marius The´riault,
Yan Kestens and Paul Villeneuve
Abstract This article is the winner of the Real Estate Valuation manuscript
prize (sponsored by The Appraisal Institute) presented at the
2001 American Real Estate Society Annual Meeting.
This hedonic study investigates the effect of landscaping on
house values, based on a detailed field survey of 760 single-
family homes sold between 1993 and 2000 on the territory of
the Quebec Urban Community. Environmental information
includes thirty-one landscaping attributes of both houses and
their immediate environment. By and large, a positive tree cover
differential between the property and its immediate
neighborhood, provided it is not excessive, translates into a
higher house value. Findings also suggest that the positive price
impact of a good tree cover in the visible surroundings is all the
more enhanced in areas with a high proportion of retired persons.
Finally, a high percentage of lawn cover as well as features such
as flower arrangements, rock plants, the presence of a hedge, etc.
all command a substantial market premium.
Objective and Context of Research
This study investigates the effect of landscaping on house values, based on a
detailed field survey of 760 single-family home sales transacted between 1993 and
2000 in the territory of the Quebec Urban Community (CUQ). The hedonic
approach is used for that purpose. While the impact of tree cover on residential
prices has already been the object of several studies, little attention has been
devoted to landscaping as such. It can be assumed that aging populations and the
ensuing propensity for ‘‘cocooning’’ should result in homeowners spending an
increasing portion of their income on landscaping. This article is an attempt to
circumscribe the phenomenon and measure the increment in value associated with
landscaping features.
140
兩
Des Rosiers, The´riault, Kestens and Villeneuve
兩
Literature Review
Over the past two decades, increasing attention has been devoted in the economic
and real estate literature to the study and measurement of the impact environmental
externalities exert on property prices (Des Rosiers, Bolduc and The´riault, 1999).
While topics encompass a wide range of issues, several authors have investigated
the effect of trees and landscaping on values. Some, like Schmitz (1988) and Yee
(1989), address that issue with respect to the office market, but most researchers
still focus on residential properties. Payne (1973) was among the first to do so.
Using traditional valuation techniques, he concludes that the market value of a
single-family house receives a 7% premium on average (between 5% and 15%)
due to arborescent vegetation, provided that there are less than thirty trees on the
lot. Beyond that point, the impact on prices is detrimental. Payne and Strom (1975)
estimate the value of seven simulated combinations of amount and distribution of
tree cover for a twelve-acre parcel of unimproved residential land in Amherst,
Massachusetts. Arrangements with trees are found to be valued 30% higher than
arrangements without ones, land price being maximized with a 67% wooded cover.
Perception studies were also performed over the past decade. Using Multiple
Listing Service-transacted suburban properties in Champaign-Urbana, Illinois,
Orland, Vining and Ebreo (1992) conducted such a study based on digitized
photographs taken from the street. Three different size-class trees were then
superimposed via video-simulation techniques. While public groups’ evaluations
show that house attractiveness is highly correlated with MLS recorded sale prices,
tree size has little effect on evaluations. While tree presence or size exerts no
impact on less expensive properties, a slight increase in value is noted for more
expensive houses when smaller trees are added, but a price decrease is associated
with larger trees. As for Kuo, Bacaicoa and Sullivan (1998), they assess the
preference pattern of 100 residents of high-rise buildings surrounding a public
open space in a densely populated neighborhood in Chicago, Illinois. Both tree
planting density and grass maintenance are tested. While the presence of trees has
strong, positive effects on residents’ preference ratings for the courtyard, grass
maintenance also has a positive impact on sense of safety, particularly when there
are fewer trees.
Finally, several hedonic analyses have been performed since the late seventies.
Combining factor analysis and multiple linear regression techniques, Morales,
Boyce and Favretti (1976) conducted a study on sixty residential sales in
Manchester, Connecticut. Four factors are used as explanatory variables, reflecting
location, house size, date of sale and tree cover, respectively. With 83% of price
variations explained by the model, the authors conclude that a good tree cover
could raise total sale price by as much as 6% to 9%. According to Seila and
Anderson (1982), newly built houses command prices that are 7% higher when
located on tree-planted lots rather than on bare ones.
Anderson and Cordell (1985) performed a first analysis on some 800 single-family
houses sold over the 1978–1980 period in Athens, Georgia. The average house
Landscaping and House Values
兩
141
JRER
兩
Vol. 23
兩
Nos. 1/2 – 2002
sold for about $47 000 and had five front-yard trees visible in its MLS
photographs. The study led to the conclusion that the presence of trees adds a 3%
to 5% premium to sale price, although other lot and building features associated
with tree cover could explain part of this increment in value, add the authors. In
a second study by Anderson and Cordell (1988) on a similar size sample involving
cheaper properties (mean sale price at $38,100), the rise in market value associated
with the presence of intermediate and large size trees stands within the 3.5% to
4.5% range, regardless of species. Broad-leaved trees contribute each roughly $376
in value, as opposed to $319 for conifers. In either study, the ensuing increase in
the city’s property tax revenues is estimated to lie between $100,000 (1988 study)
and $200,000 (1985 study) a year.
In his recent hedonic analysis on house prices in the Netherlands, Luttik (2000)
first isolates the influence of structural housing attributes on values. In a second
step, residuals from the first model are regressed against location (accessibility to
services, traffic noise, etc.) and environmental amenities (number of trees on lot,
distance to nearest green area, water body and open space). While the positive
effect of water bodies and open spaces could be demonstrated in almost every
instance, the hypothesis that a green structure commands a premium had to be
rejected in six cases out of eight. In the two cases where this variable emerges as
significant, the increment in value associated with the presence of trees or the
proximity to green areas ranges from 7% to 8%. Finally, Dombrow, Rodriguez
and Sirmans (2000) conducted a study on a sample of 269 single-family house
sales, with a mean price of $93,272. Using a semi-log functional form, a dummy
variable is included in the equation to account for the presence of mature trees.
The market-derived estimate suggests that mature trees contribute about 2% of
home values in that specific market segment.
兩
Data Bank and Analytical Approach
The Data Bank
As mentioned earlier, this study is based on a detailed field survey of 760 single-
family homes sold between 1993 and 2000 in the urbanized area of the QUC
territory. These include bungalows (one-story, detached), cottages (multi-story,
detached or semi-detached) and row houses. Conducted during the summer of
2000, the survey focuses on landscaping characteristics of homes and their
immediate environment, that is the neighborhood visible from the properties.
The overall, initial data bank includes 215 variables and factors, of which eighty-
eight physical descriptors, thirty census attributes plus twelve census factors, forty-
six location and access attributes plus two accessibility factors as well as six time
and cyclical variables. Factors are derived from previous work by Des Rosiers,
The´riault and Villeneuve (2000) whereby principal component analysis (PCA) is
performed on both 1991 census data and car travel distances and times computed
142
兩
Des Rosiers, The´riault, Kestens and Villeneuve
via the TransCAD transportation-oriented GIS software (Caliper Corp). In
addition, thirty-one environmental and landscaping attributes derived from the
field survey are added to the data bank. These are captured from the front and
side of houses and namely include trees as well as ground cover with trees—
classified by size class and type of species—flower arrangements and rock plants,
hedges, landscaped curbs, density of visible vegetation as well as roof, patio and
balcony arrangements.
The operational definition of all the variables actually used in this study—twenty-
three physical, census and access descriptors as well as eleven landscaping
attributes—is displayed in Exhibit 1. As for descriptive statistics relative to
physical, census, access and landscaping attributes, they are presented in Exhibits
2–6. While average sale price stands at around $112,000, one property sold for
$900,000 and was therefore deleted from the analysis. This results in the price
distribution being confined within the $50,000–$435,000 range. Accounting for
improvements in the housing stock, properties are, on average, sixteen years old
(apparent age), with a mean living area of roughly 120 square meters (1,300 sq.
feet). Bungalows and detached town-cottages account respectively for 42% and
40% of sales while semi-detached cottages represent 11% of the sample. As for
row houses, they account for the remaining 7%. Turning to landscaping attributes,
the average percentage of the tree cover in the neighborhood and on the property
stands at roughly 46% and 44% respectively, with the percentage of ground cover
making up for the difference. By and large, the discrepancy in the percentage of
the tree cover between the property and its neighborhood is a slight one. Quite
clearly, broad-leaved trees largely dominate in the immediate neighborhood.
Finally, 41% of houses have a hedge while landscaped curbs are present in 90%
of cases.
兩
Analytical Approach and Regression Procedures
The analysis is performed in two steps. In the first place, a basic model (Model
1) is set up using only the physical, census and access attributes of the houses.
Variable selection is done using the standard ‘‘enter’’ procedure, which was
validated via a stepwise approach. With respect to census and access attributes,
both individual variables and PCA-derived factors are successively tested. While
the latter allow for a more qualified interpretation of the urban dynamics
underlying the price determination process than the former do, they do not lead
to better model performances. In this article therefore, individual census and
access variables are used instead of factors.
Once Model 1 parameters are stabilized, they are forced into Models 2 and 3,
which also include landscaping attributes. Again, both standard and stepwise
regression procedures are successively applied to the latter for final variable
selection. Several combinations of landscaping arrangements were tested and
various mathematical transformations performed on variables. By and large,
Landscaping and House Values
兩
143
JRER
兩
Vol. 23
兩
Nos. 1/2 – 2002
Exhibit 1 兩 Operational Definition of Variables
Variable Attributes Type
BUNGALOW One-story, single-family detached house D
COTTAGE More than one-story, single-family detached townhouse D
SEMIDET More than one-story, single-family semi-detached townhouse D
ROW Row house D
APPAGE Apparent age of the property, in years M
LIVAREA Living area of the property, in square meters M
LNLOTSIZ Natural logarithm of the lot size, expressed in square meters M
QUALINF Indicates a below-average overall building quality D
QUALSUP Indicates an above-average overall building quality D
SUPFLOOR Superior quality, hardwood floor D
ATTGARAG Attached garage D
DETGARAG Detached garage D
EXCAPOOL Excavated pool D
BASEFINH Finished basement D
OVEN Built-in oven in the kitchen (modern kitchen) D
%AGE45 64 % of individuals aged 45-64 M
%AGE65 UP % of individuals aged 65 and over M
%WOMEN % of women M
%SGLHLD % of one-person households M
%DW46 60 % of buildings built between 1946 and 1960 M
%UNIVDEGR % of individuals with a university degree M
ULAVLCTM Car travel time from property to Laval University, in minutes M
HIGHW1KM The property is within one km. from a highway exit D
%Tree Prop Percentage of tree cover on the property M
%Tree Nbhd Percentage of tree cover in the neighborhood M
%Grnd Prop Percentage of ground cover on the property M
%Grnd Nbhd Percentage of ground cover in the neighborhood M
Prop-Nbhd
%Tree
Difference in the % of tree cover between the property and the
neighborhood
M
Prop Nbhd
%Tree Ratio
Property vs. neighborhood % tree cover ratio M
%Brdlvd Nbhd Percentage of broad-leaved trees in the neighborhood M
Density Visible
Veg
Density of vegetation visible from the property R
Hedge Presence of a landscaped hedge /wall D
Patio Presence of a landscaped patio D
Curbs Presence of landscaped curbs D
Notes: N.B.: M Metric; D Dummy; R Rank.
144
兩
Des Rosiers, The´riault, Kestens and Villeneuve
Exhibit 2 兩 Descriptive Statistics—Continuous Descriptors
SPRICE ($) LOTSIZE APPAGE AGE LIVAREA %AGE45 64 %AGE65 UP %PERSHLD %WOMEN %SGLHLD %UNIVDEGF
Mean 112,096 659 16 19 123 20.3 6.7 2.8 50.9 14.8 23.1
Median 92,500 572 15 15 109 20.1 4.6 3.0 50.5 11.6 18.7
Mode 82,000 372 0 0 86 20.1 3.8 3.0 50.2 6.0 14.7
Std. Dev. 61,488 816 13 20 49 7.8 6.7 0.4 2.4 11.0 13.9
Min. 50,000 125 0 0 39 6.9 0 1.3 37.5 0 2.1
Max. 900,000 18,767 54 164 627 37.9 53.7 3.5 65.4 78.0 65.1
Note: N 76.
Landscaping and House Values
兩
145
JRER
兩
Vol. 23
兩
Nos. 1/2 – 2002
Exhibit 3 兩 Descriptive Statistics—Dummy Variables
BUNGALOW COTTAGE SEMIDET ROW QUALINF QUALINF QUALSUP EXCAPOOL ATTGARAG DETGARAG
N Valid 760 760 760 760 760 760 760 760 760 760
N Missing 0 0 0 0 0 0 0 0 0 0
Mean 0.42 0.40 0.11 0.07 0.07 0.01 0.05 0.07 0.10 0.15
146
兩
Des Rosiers, The´riault, Kestens and Villeneuve
Exhibit 4 兩 Descriptive Statistics—Landscaping Variables
Density
Visible Veg
%Tree
Nbhd
%Tree
Prop
%Grnd
Prop
%Grnd
Nbhd
Prop Nbhd
%Tree
Prop Nbhd
%Tree Ratio Hedge Patio Curbs
%Brdlvd
Nbhd
Mean 2.24 45.6 43.6 55.4 53.7 2.0 0.97 0.41 0.01 0.90 84.7
Median 2.00 40.0 40.0 60.0 60.0 0 0.97 0 0 1.00 90.0
Mode 2 40.0 30.0 70.0 60.0 0 0.97 0 0 1.00 90.0
Std. Dev. 0.53 21.9 25.8 26.1 22.0 19.1 0.58 0.49 0.11 0.30 15.5
Min. 0 0 0 0 0 80.0 0 0 0 0 0
Max. 4 90.0 100.0 100.0 100.0 70.0 6.4 1.0 1.0 1.0 100.0
Note: N 76.
Landscaping and House Values
兩
147
JRER
兩
Vol. 23
兩
Nos. 1/2 – 2002
Exhibit 5 兩 Descriptive Statistics—Interactive Variables
%Tree
Nbhd*%Age65 up
%Grnd
Prop*Cottage
%Grnd
Prop*Bungalow
Prop–Nbhd%Tree
ctd*%age45 65
%Grnd Prop
ctd*Cottage
%Tree Nbhd
ctd*%Women ctd
Mean 3.44 0.21 0.25 0.03 0.87 0.13
Median 1.60 0 0 0 0 0.05
Mode 0.26
a
0 0 0 0 0.11
Std. Dev. 4.45 0.31 0.33 1.51 15.97 0.58
Min. 0 0 0 8.95 55.37 1.78
Max. 35.82 1.00 1.00 9.30 44.63 6.10
Notes: N 76.
a
Multiple modes exist. The smallest value is shown.
148
兩
Des Rosiers, The´riault, Kestens and Villeneuve
Exhibit 6 兩 ANOVA Results
Model
Sum of
Squares df
Mean
Square F Sig.
1 Regression 93.451 18 5.192 264.601 0.000
Residual 14.520 740 1.962E-02
Total 107.971 758
2 Regression 93.846 19 4.939 258.425 0.000
Residual 14.125 739 1.911E-02
Total 107.971 758
3 Regression 93.215 24 3.884 193.206 0.000
Residual 14.755 734 0.020
Total 107.971 758
Note: The dependent variable LNSPRICE.
interactive variables (Jaccard, Turrisis and Wan, 1990) proved to be particularly
efficient at capturing household behavior with respect to landscaping. While
absolute interactions are resorted to in Model 2, relative interaction variables are
also used in Model 3, each continuous component of the resulting descriptors
being first centered, thereby reflecting its departure from the mean.
Both the linear and semi-log functional forms were tested, with the latter yielding
much better overall performances and, by and large, higher ‘‘t’’ values. Moreover,
regression coefficients derived from the semi-log form are expressed as
relative—rather than absolute—implicit prices, thereby allowing for a more
flexible interpretation of the contribution of housing attributes to property value.
The general formulation of the final hedonic equation underlying the current
empirical investigation can thus be expressed as follows:
BB B B B
01iPhys 2iCensus 3i.Access 4iLandsc
Y eeeeee, (1)
** * * *
Where Y is the sale price while Phys, Census, Access and Landsc represent the
four series of descriptors used in the analysis. This, in turn, can be put as:
LnY B B Phys BCensus B
** *
01i 2i 3i
Access B Landsc . (2)
*
4i
Landscaping and House Values
兩
149
JRER
兩
Vol. 23
兩
Nos. 1/2 – 2002
Exhibit 7 兩 Regression Results—Model 1/Basic Attributes
Unstandardized
Coefficients
Standardized Coefficients
Collinearity
Statistics
B Std. Error Beta t Sig. VIF
Constant 10.546 0.089 117.83 0.000
LIVAREA 0.004 0.000 0.447 22.84 0.000 2.11
LNLOTSIZ 0.106 0.014 0.130 7.54 0.000 1.64
ROW 0.158 0.024 0.105 6.64 0.000 1.37
SEMIDET 0.104 0.018 0.085 5.72 0.000 1.23
QUALINF 0.224 0.045 0.071 4.98 0.000 1.12
QUALSUP 0.012 0.029 0.066 4.03 0.000 1.46
APPAGE 0.042 0.001 0.394 19.32 0.000 2.29
SUPFLOOR 0.042 0.011 0.055 3.73 0.000 1.20
ATTGARAG 0.116 0.019 0.090 5.98 0.000 1.26
SETGARAG 0.072 0.016 0.069 4.48 0.000 1.30
EXCAPOOL 0.069 0.023 0.045 3.01 0.003 1.26
BASEFINH 0.055 0.011 0.072 5.01 0.000 1.15
OVEN 0.066 0.015 0.067 4.52 0.000 1.19
%AGE65 UP 0.006 0.001 0.108 5.37 0.000 2.24
%DW46 60 0.002 0.000 0.059 3.07 0.002 2.04
%UNIVDEGR 0.007 0.000 0.245 13.47 0.000 1.83
HIGHW1KM 0.039 0.013 0.050 3.12 0.002 1.42
ULAVLCTM 0.017 0.002 0.209 9.10 0.000 2.90
Notes: The dependent variable is LNSPRICE. R
2
.866; Adj. R
2
.862; F-value 264.6 (.000);
and the std. error of the estimate 0.1401 (15.04%).
Finally, no extreme outliers were filtered out from the study since nothing could
lead to the conclusion that these were not representative of the residential market
under analysis.
兩
Summary of Major Findings
The Basic Model
As can be seen from Exhibit 7, the overall performances of Model 1 are quite
good. With thirteen physical, three census and two access variables in the equation,
150
兩
Des Rosiers, The´riault, Kestens and Villeneuve
Exhibit 8 兩 Regression Results—Model 2/Landscaping Absolute Interaction Variables
Unstandardized
Coefficients Unit Adjustment Factors
Collinearity
Statistics
B Std. Error
Exp
(B) t Sig. VIF
Constant 10.369 0.076 135.99 0.000
LIVAREA 0.004 0.000 1.004 20.88 0.000 2.56
LNLOTSIZ 0.105 0.013 7.94 0.000 1.48
QUALINF 0.245 0.045 0.783 5.49 0.000 1.13
QUALSUP 0.131 0.028 1.140 4.60 0.000 1.49
APPAGE 0.011 0.001 0.989 18.43 0.000 2.26
SUPFLOOR 0.039 0.011 1.040 3.49 0.001 1.22
ATTGARAG 0.109 0.019 1.116 5.63 0.000 1.30
SETGARAG 0.072 0.016 1.074 4.46 0.000 1.32
EXCAPOOL 0.081 0.023 1.084 3.56 0.000 1.25
BASEFINH 0.050 0.011 1.051 4.54 0.000 1.18
OVEN 0.068 0.014 1.070 4.69 0.000 1.20
%UNIVDEGR 0.007 0.000 1.007 14.64 0.000 1.79
ULAVLCTM 0.141 0.002 0.986 8.48 0.000 2.37
%DW46 60 0.001 0.000 1.001 2.88 0.004 2.03
%Tree Nbhd*%Age65 up 0.011 0.002 1.011 6.74 0.000 2.17
%Grnd Prop*Bungalow 0.002 0.000 1.002 8.73 0.000 2.19
%Grnd Prop*Cottagep 0.002 0.000 1.002 7.54 0.000 2.43
Prop – Nbhd%Tree 0.002 0.000 1.002 5.50 0.000 1.40
Hedge 0.039 0.111 1.039 3.51 0.000 1.16
Notes: The dependent variable is LNSPRICE. R
2
.869; Adj. R
2
.866; F-value 258.4 (.000);
and the std. error of the estimate 0.1382 (14.82%).
the adjusted R
2
is .86 while the F-value reaches 265. In spite of a relatively high
prediction error (15%), which stems from both the regional scope of the model
and the wide range of sale prices, all regression coefficients are consistent in sign
and magnitude with theoretical expectations and statistically significant at the .01
level. Particularly noteworthy is the marginal contribution of the education/income
variable (%UNIVDEGR) that comes next to property age (APPAGE) in terms of
t-value as well as that of the main access descriptor (ULAVALCTM), which
confirms the strategic role played by Laval University as a regional activity center.
While transactions spread over an eight-year period, the model retains no time or
Landscaping and House Values
兩
151
JRER
兩
Vol. 23
兩
Nos. 1/2 – 2002
cyclical variable, as prices remained relatively stable until late 1997 and recovered
only slightly thereafter. Finally, multicollinearity remains within very acceptable
limits, with the highest VIF standing at 2.90.
Landscaping Features—Absolute Interactions
While inserting landscaping variables in the hedonic equation only slightly
improves overall explanatory and predictive model performances (Adj. R
2
up to
.87 and SEE down to 14.8%), it does not modify substantially the implicit prices
of the basic housing attributes remaining in the equation—although four of these
(ROW, SEMIDET, ‰AGE65
UP and HIGHW1KM) are rejected from the initial
model. Most of all, Model 2 (Exhibit 8) provides a clear indication that
landscaping features do exert a significant impact on property prices. While all
five related descriptors that enter the model emerge as highly significant without
causing undue collinearity, three of these—showing the strongest t values—are
actually interactive variables that bring out absolute interactions between
landscaping features on the one hand and property type or demographic structure
on the other. Such a device proves very useful in that it allows for a better qualified
interpretation of landscaping influence on values.
The findings can be summarized as follows:
䡲 A positive differential in the percentage of tree cover between the
property and its immediate neighborhood (Prop-Nbhd %Tree) raises
house value by roughly 0.2% for each percentage point, which could be
interpreted as a ‘‘scarcity’’ premium;
䡲 The higher the proportion of retired people in the neighborhood, the more
beneficial to the market value of a given property the presence of trees
in its vicinity is (%Tree Nbhd
*
%Age65 up);
䡲 For bungalows and cottages though, the higher the percentage of ground
cover (lawn, flower arrangements, rock plants, etc.) on the property, the
higher the value (%GrndProp
*
Bungalow, %GrndProp
*
Cottage), each
percentage point adding some 0.2% to the price; and
䡲 Finally, the presence of a hedge or landscaped wall (Hedge) raises a
property’s value by nearly 4%, which can be assumed to mirror both the
enhanced visual appearance from the home and the increased intimacy it
provides.
By and large, the results obtained with Model 2 are in line with the literature.
Applying the mean value for each landscaping-related variable and assuming the
presence of a hedge results in a 7.7% market premium for either a typical
bungalow or cottage. This is quite similar to findings by Morales, Boyce and
Favretti (1976—6% to 9%), Seila and Anderson (1982—7%) and, more recently,
Luttik (2000—7% to 8%).
152
兩
Des Rosiers, The´riault, Kestens and Villeneuve
兩
Landscaping Features—Relative Interactions
Model 3 (Exhibit 9) mainly differs from Model 2 in that several of the landscaping
interactions tested are based on centered, rather than original, variables. Performed
on continuous descriptors only, this procedure generates complex implicit prices
reflecting the impact of a departure from the local standard. Relative interactions
between landscaping features and other price determinants are thus measured. At
first glance, Model 3 displays slightly lower overall performances compared to
Model 2, mainly with respect to the F-value, which drops to 193.2, from 258.4
previously. With twenty-four descriptors emerging as significant (ROW, SEMIDET
and ‰AGE65
UP re-enter the model while ULAVLCTM is rejected), several
landscaping features—used as single characteristics or in interaction—are shown
to affect prices.
Thus:
䡲 With a t-value in excess of 5 and a positively signed coefficient, the
property vs. neighborhood percentage-of-tree-cover ratio (Prop
Nbhd
%Tree Ratio) corroborates Model 2 findings with regard to the ‘‘scarcity’’
premium (in this case, 7.3% per measurement unit) assigned to houses
with more trees than surrounding properties.
䡲 In contrast, a negative adjustment is required in neighborhoods where
early boomers (i.e., people aged 45–64) dominate (Prop-Nbhd %Tree
*
Age45 64).
䡲 An increment is also added to house value wherever the proportion of
women in the residential area and the tree cover in the immediate vicinity
of the property are both either above or below the local average; in
contrast, prices drop if the two components evolve in opposite directions
(%Tree Nbhd ctd
*
%Women ctd).
䡲 As with Model 2, cottages experience a rise in their market value (0.1%)
for each additional positive departure-from-mean percentage point of
ground cover on the property, and a drop if the departure is negative
(%GrndProp ctd * Cottage).
䡲 Quite interestingly, the density of the vegetation visible from the property
(Density Visible Veg) impacts negatively on prices, each rank unit
resulting in a loss of roughly 2.2% of value.
䡲 Considering the distribution of this variable, such a finding can be
interpreted as a confirmation of Payne’s (1973) conclusions regarding
excessive tree cover.
Finally, a landscaped patio, a hedge as well as landscaped curbs add respectively
12.4%, 3.6% and 4.4% to the market value of a house, respectively.
While Model 3 provides a great variety of landscaping influences on property
values, pertaining regression coefficients display, by and large, lower t-values than
Landscaping and House Values
兩
153
JRER
兩
Vol. 23
兩
Nos. 1/2 – 2002
Exhibit 9 兩 Regression Results—Model 3/Landscaping Relative Interaction Variables
Unstandardized
Coefficients Unit Adjustment Factors
Collinearity
Statistics
B Std. Error
Exp
(B) t Sig. VIF
Constant 10.316 0.091 113.52 0.000
LIVAREA 0.004 0.000 1.004 21.56 0.000 2.25
LNLOTSIZ 0.081 0.015 5.52 0.000 1.75
ROW 0.141 0.025 0.868 5.76 0.000 1.42
SEMIDET 0.121 0.019 0.886 6.47 0.000 1.26
QUALINF 0.238 0.046 0.788 5.18 0.000 1.14
QUALSUP 0.157 0.029 1.170 5.43 0.000 1.47
APPAGE 0.010 0.001 0.990 16.64 0.000 2.34
SUPFLOOR 0.026 0.011 1.027 2.29 0.022 1.24
ATTGARAG 0.117 0.020 1.124 5.93 0.000 1.28
SETGARAG 0.080 0.016 1.083 4.88 0.000 1.31
EXCAPOOL 0.093 0.023 1.097 3.98 0.000 1.27
BASEFINH 0.048 0.011 1.050 4.33 0.000 1.17
OVEN 0.079 0.015 1.082 5.33 0.000 1.21
%AGE65 up 0.008 0.001 1.008 6.76 0.000 2.20
%DW46 60 0.002 0.000 1.002 3.61 0.000 2.11
%UNIVDEGR 0.009 0.000 1.009 20.81 0.000 1.36
Prop – Nbhd%Tree
Ratio
0.071 0.014 1.073 5.11 0.000 2.43
Prop – Nbhd
%Tree*%Age45 64
0.006 0.002 0.994 3.46 0.001 2.45
%Grnd Prop
ctd*Cottage
0.001 0.000 1.001 2.93 0.003 1.31
%Tree Nbhd
ctd*%Women ctd
0.036 0.010 1.036 3.56 0.000 1.25
Density visible veg 0.023 0.010 0.978 2.30 0.022 1.20
Hedge 0.036 0.011 1.036 3.16 0.002 1.15
Patio 0.117 0.048 1.124 2.41 0.016 1.04
Curbs 0.043 0.018 1.044 2.36 0.019 1.13
Notes: The dependent variable is LNSPRICE. R
2
.863; Adj. R
2
.859; F-value 193.2 (.000);
and the std. error of the estimate 0.1418 (15.23%).
154
兩
Des Rosiers, The´riault, Kestens and Villeneuve
is the case with Model 2 and are therefore less reliable. Having said that, no
inconsistency can be detected between the two models, each of which throwing a
different shade on the phenomenon under analysis and complementing the other.
兩
Testing for Spatial Autocorrelation and
Heteroscedasticity of Residuals
Implicit prices derived from hedonic modeling may not be considered as reliable
unless it can be shown that model residuals are both exempt from any significant
spatial autocorrelation and homoscedastic (Dubin, 1988; Anselin and Rey, 1991;
Ord and Getis, 1995; Can and Megbolugbe, 1997; Basu and Thibodeau, 1998;
Pace, Barry and Sirmans, 1998; and Des Rosiers and The´riault, 1999). Indeed,
either phenomenon is highly detrimental to the efficiency of the statistical tests
used to assess the statistical significance of ordinary least squares regression
coefficients (Anselin 1990). It is worth mentioning that while structural
heteroscedasticity can occur in the absence of any spatial autocorrelation, the
reverse is not true, the latter causing the former.
兩
Measuring Spatial Autocorrelation Using Moran’s I
Named after Moran (1950), the Moran’s I is used to measure spatial
autocorrelation in the models residuals. Considering (x
i
,y
i
,z
i
), a data triad, where
x and y are projected Euclidean co-ordinates of a point location (i), and z is any
numerical value associated to that location, one can define indexed vectors
containing values measured at n locations. The Euclidean distance (d
ij
) between
any pair of data at locations i and j is computed using their Cartesian co-ordinates
(x
i
, y
i
) and (x
j
, y
j
):
22
d 兹(x x ) (y y ) . (3)
ij i j i j
The mathematical average of the n values of vector z is noted:
n
z z /n. (4)
冘
i
i
1
The Moran’s I index measures autocorrelation between values of the z vector,
considering some weight (w
ij
) that is a function of spatial proximity between any
(i and j) pair of observations:
Landscaping and House Values
兩
155
JRER
兩
Vol. 23
兩
Nos. 1/2 – 2002
nn,j
i
w (z z)(z z)
冘冘
ij i j
n
i
1 j
1
I , and (5)
nn,j
1n
2
w (z z)
冘冘 冘
ij i
i
1 j
1 i
1
a
p
ij
w , (6)
ij
2
d
ij
where p
ij
measures some degree of interaction between i and j. The exponent (a)
is chosen to express the relative importance of the interaction links between the
n by n
1 pairs of observations. For the purpose of this article, was kept equal
a
p
ij
to one, meaning that interaction between each property was weighted only by the
squared inverse distance between them. This clearly expresses the decrease in
influence under gravitational effect.
The sampling distribution of the mean E(I) and variance Va r(I) expectations of
Moran’s I are known. They form the basis of a parametric test for confirming the
significance of experimental results. Knowing that the theoretical distribution of
Moran’s I is gaussian, the standard normal deviate Z(I) between the measure and
its expectation is used in order to reject the null hypothesis that the observed
autocorrelation occur by chance only (Odland, 1988). We thus have:
I E(I)
Z(I) , where (7)
兹VAR(I)
1
E(I) , and (8)
(n
1)
2
nn,j
i
nS S 3 w
冘冘
冉冊
21
n 2 ij
i
1 j
1
VAR(I) , with (9)
2
nn,j
i
w (n 1)
冘冘
冉冊
ij 2
i
1 j
1
2
nn,j
inn,j
in,j
i
1
2
S (w w ) and S w w .
冘冘 冘 冘 冘
冉冊
1 ij ji 2 ij ji
2
i
1 j
1 i
1 j
1 j
1
(10)
Spatial autocorrelation indexes reported in this article are computed using the
MapStat software developed by The´riault and take into account all pairs of
156
兩
Des Rosiers, The´riault, Kestens and Villeneuve
Exhibit 10 兩 Testing for Spatial Autocorrelation of Models Residuals
transactions linking homes located within a radius of 1,500 meters (roughly one
mile) from one another (Euclidean distances). Moran’s I are computed on both
sale price and Models 1 to 3 residuals. Resulting correlograms as well as the
overall autocorrelation index for properties located within a 1,500 meters radius
are shown in Exhibit 10. As expected, sale prices display a high degree (0.9) of
spatial dependence in the immediate vicinity of a residence, with the Moran’s I
Landscaping and House Values
兩
157
JRER
兩
Vol. 23
兩
Nos. 1/2 – 2002
Exhibit 11 兩 Goldfeld-Quandt Test: Ratio of Upper to Lower Segment
Ranking
Variable Group 1 Group 2 F Prob.
LnSPRICE 1,876 4,148 2,211 0,000
dl 228 227
APPAGE 2,752 5,522 2,007 0,000
dl 227 226
LIVAREA 3,052 6,133 2,010 0,000
dl 227 226
tending to fall rapidly with distance and stabilizing at around 0.5 beyond 450
meters. While models 1 and 2 residuals are still flawed by a significant degree of
spatial autocorrelation within a 750 meters radius, even after socio-economic,
access and landscaping absolute interaction variables are introduced, resorting to
relative interactions between landscaping features and other price determinants
greatly improves the picture. Indeed, nearby autocorrelation, which is most
detrimental to the stability of implicit prices of housing attributes, is no more
significant. As can be seen from the lower portion of Exhibit 10, the overall index
measured over a 1,500 meters radius sharply falls from roughly 0.16 and 0.15 in
Models 1 and 2 respectively to 0.05 in Model 3 (prob. 0.2545).
兩
Measuring Heteroscedasticity Using Glejser’s Test
With spatial autocorrelation under control, it is now relevant to test for the
presence of heteroscedastic residuals. The heteroscedasticity issue was dealt with
several decades ago by Goldfeld and Quandt (1965) and Glejser (1969) who have
designed statistical tests for its detection. According to the former, G-Q test, the
global sales sample is first split into a lower and an upper portion segmented
along one or several criteria (sale price, age, living area, etc.), the middle 20%
segment of the distribution being left aside. Either submarket is then applied the
initial model’s parameters and the resulting squared residuals are summed up. A
higher and a lower value are thus derived and their ratio, which follows a Fisher
distribution, is finally computed: where statistically significant, this ratio indicates
that model residuals are not homoscedastic. As for the latter, Glejser test, it simply
consists in successively regressing selected variables against the absolute value of
model residuals, using various functional forms. Heteroscedasticity is said to be
present in the residuals if any b
i
coefficient emerges as statistically significant.
Results obtained with both tests are displayed in Exhibits 11 and 12. Both tests
are conclusive and clearly indicate the presence of heteroscedasticity in the
residuals with respect to sale price, apparent age and living area. Considering that
158
兩
Des Rosiers, The´riault, Kestens and Villeneuve
Exhibit 12 兩 Glejser Test: Regression on Absolute Value of Residuals
Reference
Variable
Functional
Form Coefficient t Prob. Adj. R
2
LnSPRICE Linear b1 4.06 0.0001 0.020
Inverse b1 3.81 0.0002 0.018
Quadratic b1 4.96 0.0000 0.051
b2 5.06 0.0000
APPAGE Linear b1 4.38 0.0000 0.023
Quadratic b1 1.88 0.0608 0.023
b2 0.54 0.5892
LIVAREA Linear b1 5.19 0.0000 0.033
Inverse b1 4.11 0.0000 0.206
Quadratic b1 2.06 0.0396 0.033
b2 0.81 0.4158
several spatial dimensions are already accounted for in the modeling process, it
can be assumed that a-spatial, structural heteroscedasticity is at stake. Furthermore,
the very wide range of sale prices ($50,000–$435,000) on which this study is
based may largely explain the scope of the problem encountered.
兩
Conclusion
This study investigated the effect of landscaping on house values, based on a
detailed field survey of 760 single-family home sales transacted between 1993 and
2000 in the territory of the Quebec Urban Community, using for that purpose the
hedonic approach. Conducted during the summer of 2000, this survey focuses on
landscaping characteristics of homes and their immediate environment that is the
adjacent neighborhood visible from the transacted properties. Environmental
information was captured from the front and side of houses and includes thirty-
one attributes dealing with tree as well as ground cover—with trees being
classified by size class and type of species—, flower arrangements and rock plants,
hedges, landscaped curbs, density of visible vegetation as well as roof, patio and
balcony arrangements. Landscaping features are added to an array of physical,
census and access attributes. Sale prices range from a minimum of $50,000 to a
maximum of $435,000, with the mean price standing at $112,000. Property types
include bungalows—one-story, detached (42% of sample)—, cottages—multi-
story, detached (40%)or semi-detached (40%)—and row houses (7%).
Once the basic model (Model 1) is calibrated using the physical, census and access
characteristics of properties, landscaping features are added to the hedonic
equation, with both individual attributes and interactive variables being used.
While absolute interactions are resorted to in Model 2, relative interactions based
Landscaping and House Values
兩
159
JRER
兩
Vol. 23
兩
Nos. 1/2 – 2002
on centered variables are used in Model 3. As fewer significant landscaping-related
coefficients are derived from the former model, they in turn display greater
stability than those generated by the latter. Having said that, each model brings
complementary information and remains consistent with one another.
By and large, a positive tree cover differential—or a more-than-unity
ratio—between the property and its immediate neighborhood translates into a
higher house value, although a negative adjustment is required where early
boomers—aged 45-64—dominate (Model 3). While the relative importance of tree
cover in the visible surroundings also exerts a positive impact on property prices,
the effect is all the more enhanced in areas with a high proportion of retired
persons (Model 2). If trees seem to be valued by most homeowners, a high
percentage of ground cover (lawn, flower arrangements, rock plants, etc.) also
commands a market premium in the case of bungalows and cottages (Model 2);
moreover, the price of cottages benefits from an above-average ground cover
whereas a below-average one is detrimental (Model 3). Quite interestingly, an
above-average density of the vegetation visible from the property impacts
negatively on prices (Model 3), in line with Payne’s (1973) conclusions regarding
excessive tree cover. Finally, a hedge, a landscaped patio as well as landscaped
curbs all command a substantial market premium: while it amounts to between
3.6% (Model 3) and 3.9% (Model 2) of property value for a hedge, it reaches
12.4% in the case of a patio and 4.4% for landscaped curbs (Model 3). Applying
Model 2 using the mean value for each landscaping-related variable and assuming
the presence of a hedge results in a 7.7% market premium for either a typical
bungalow or a cottage.
In conclusion, research findings are in line with the literature on the subject. They
suggest that using a parametric estimation technique can yield reliable, space-
specific hedonic prices if landscaping interactive variables are resorted to.
Furthermore, using relative interaction variables substantially reduces the
detrimental effects spatial autocorrelation exerts on the stability of regression
parameters.
Conclusive though it might be, the current study has some limitations that deserve
further research. First, it does not account for possible landscaping improvements
carried out between the transaction date and the survey period (Summer 2000),
which could distort hedonic prices. Second, the current data bank does not allow
an investigation into the links between, on the one hand, homeowners’ preferences
for landscaping features and, on the other hand, their socio-demographic and
economic profile. An extensive phone survey is presently underway, which should
soon palliate these informational flaws and lead to an even more reliable
assessment of how landscaping shapes house values.
兩
References
Anderson, L. M. and H. K. Cordell, Residential Property Values Improved by Landscaping
with Trees, Southern Journal of Applied Forestry, 1985, 9, 162–66.
160
兩
Des Rosiers, The´riault, Kestens and Villeneuve
——., Influence of Trees on Residential Property Values in Athens, Georgia (U.S.A.): A
Survey Based on Actual Sales Prices, Landscape and Urban Planning, 1988, 15, 153–64.
Anselin, L., Spatial Dependence and Spatial Structural Instability in Applied Regression
Analysis, 1990, Journal of Regional Science, 30:2, 185–207.
Anselin, L. and S. Rey, Properties of Tests for Spatial Dependence in Linear Regression
Models, Geographical Analysis, 1991, 23:2, 112–31.
Basu, S., and T. G. Thibodeau, Analysis of Spatial Autocorrelation in House Prices, Journal
of Real Estate Finance and Economics, 1998, 17:1, 61–86.
Can, A. and I. Megbolugbe, Spatial Dependence and House Price Index Construction,
Journal of Real Estate Finance and Economics, 1997, 14, 203–22.
Des Rosiers, F., A. Bolduc and M. The´riault, Environment and Value: Does Drinking Water
Quality Affect House Prices?, Journal of Property Investment and Finance, 1999, 17:5,
444–63.
Des Rosiers, F. and M. The´riault, House Prices and Spatial Dependence: Towards an
Integrated Procedure to Model Neighborhood Dynamics, Working Paper # 1999-002,
Faculty of Business Administration, Laval University, January 1999.
Des Rosiers, F., M. The´riault and P.-Y. Villeneuve, Sorting out Access and Neighbourhood
Factors in Hedonic Price Modelling, Journal of Property Investment and Finance, 2000,
18:3, 291–315.
Dombrow, J., M. Rodriguez and C. F. Sirmans, The Market Value of Mature Trees in
Single-Family Housing Markets, The Appraisal Journal, 2000, 68, 39–43.
Dubin, R. A., Estimation of Regression Coefficients in the Presence of Spatially
Autocorrelated Error Terms, Review of Economics and Statistics, 1988, 70, 466–74.
Glejser, H., A New Test for Heteroscedasticity, Journal of the American Statistical
Association, 1969, 64, 316–23.
Goldfeld, S. M. and R. E. Quandt, Some Tests for Homoscedasticity, Journal of the
American Statistical Association, 1965, 60, 539–47.
Jaccard, J., R. Turrisi and C. K. Wan, Interaction Effects in Multiple Regression, Sage
University Papers, Series Quantitative Applications in the Social Sciences, No. 72, Sage
Publications, Newbury Park, CA, 1990.
Kuo, F. E., M. Bacaicoa and W. C. Sullivan, Transforming Inner-City Landscapes: Trees,
Sense of Safety and Preferences, Environment and Behavior, 1998, 30:1, 28–59.
Luttik, J., The Value of Trees, Water and Open Space as Reflected by House Prices in the
Netherlands, Landscape and Urban Planning, 2000, 48, 161–67.
Morales, D., B. N. Boyce and R. J. Favretti, The Contribution of Trees to Residential
Property Value: Manchester, Connecticut, Valuation, 1976, 23:2, 26–43.
Moran, P. A. P., Notes on Continuous Stochastic Phenomena, Biometrika, 1950, 37, 17–
23.
Odland, J., Spatial Autocorrelation, Scientific Geography, Vol. 9, Sage Publications,
Newbury Park, 1988.
Ord, J. K. and A. Getis, Spatial Autocorrelation Statistics: Distribution Issues and an
Application, Geographical Analysis, 1995, 27:4, 286–306.
Orland, B., J. Vining and A. Ebreo, The Effect of Street Trees on Perceived Values of
Residential Property, Environment and Behavior, 1992, 24:3, 298–325.
Pace, R. K., R. Barry and C. F. Sirmans, Spatial Statistics and Real Estate, Journal of Real
Estate Finance and Economics, 1998, 17:1, 5–14.
Landscaping and House Values
兩
161
JRER
兩
Vol. 23
兩
Nos. 1/2 – 2002
Payne, B. R., The Twenty-Nine Tree Home Improvement Plan, Natural History, 1973, 82,
74–5.
Payne, B. R. and S. Strom, The Contribution of Trees to the Appraised Value of
Unimproved Residential Land, Valuation, 1975, 22:2, 36–45.
Schmitz, W. G., Amenities: The Name of the Game in Office Space, Commercial Investment
Real Estate Journal, Chicago, May/June 1988.
Seila, A. F. and L. M. Anderson, Estimating Costs of Tree Preservation on Residential
Lots, Journal of Arboriculture, 1982, 8, 182–85.
Yee, J., Suburban Office Landscaping, The Journal of Real Estate Development, Boston,
Winter 1989.
The research was funded by the Social Sciences and Humanities Research Council of
Canada and by the FCAR Research Fund (Quebec). The authors thank Quebec City’s
Appraisal Division and its director, Jean-Guy Kirouac, as well as Jean Guyard, for
having provided the data. The authors also thank Paul James, Mireille Campagna
and the reviewers.
Franc¸ois Des Rosiers, Laval University, Canada or Francois.Desrosiers@fsa.
ulaval.ca.
Marius The´riault, Laval University, Canada or Marius.Theriault@crad.ulaval.ca.
Yan Kestens, Laval University, Canada or Yan.Kestens@crad.ulaval.ca.
Paul Villeneuve, Laval University, Canada or Paul.Villeneuve@crad.ulaval.ca.
