HomeMy WebLinkAboutJensen-et-al-Wind-Turbines-Property-Values.pdf
The Vindication of Don Quijote:
The impact of noise and visual pollution
from wind turbines on local residents
in Denmark
Cathrine Ulla Jensen
Toke Emil Panduro
Thomas Hedemark Lundhede
2013 / 13
Title:
The Vindication of Don Quijote: The impact of noise and visual pollution from wind turbines
on local residents in Denmark.
Authors:
Cathrine Ulla Jensena, Toke Emil Panduro & Thomas Hedemark Lundhede
Affiliation:
Department of Food and Resource Economics, Faculty of Sciences, University of
Copenhagen
a) Corresponding Author:
Cathrine Ulla Jensen
Email: CathrineUJensen@gmail.com,
Tel: +45 2149 7784
Abstract
In this article we quantify the marginal external effects of nearby land based wind turbines on
property prices capitalized through traded residential properties located within 2,500 meters
or less. We succeed in separating the effect of noise and visual pollution from wind turbines.
This was achieved by using a dataset covering 21 municipalities and consisting of 12,640
traded residential properties sold in the period 2000-2011. We model the hedonic price
function in two steps. First we detrend data across municipalities using a pooled cross
sectional model which allows for different price trends across municipalities. Second we
control for spatial autocorrelation by using explicit spatial models. Properties affected by
noise and visual pollution from wind turbines are identified using Geographical Information
Systems. Our results show that wind turbines have a significant negative impact on the price
schedule of neighboring residential properties. The visual pollution accounts for 3.15% of the
residential sales price. The price premium declines with distance by about 0.242% of the
sales price for every 100 meters. The effect of noise depends on the noise level emitted and
ranges from 3% to 7% of the sale price for residential properties.
Keywords:
Valuation, wind turbines, spatial autocorrelation, hedonic house price modeling
Highlights
We estimate the negative impact of visual and noise pollution from wind turbines.
We construct viewsheds based high resolution digital surface model.
We estimate the hedonic price model using spatial autoregressive models.
1
1. Introduction
In the 16th century, the fictional character Don Quijote thought that windmills were alien to
the landscape. Many people have similar views about wind turbines today. The installation of
land based wind turbines is controversial and is often met with opposition from the local
community (Wolsink 2000), which often takes the form of a “Not in my back yard” argument.
The general need to increase renewable energy, and install wind turbines in particular, is
acknowledged, but at the same time the location of local wind turbine projects is opposed.
The noise and visual appearance of wind turbines makes them very unattractive neighbors
(Devine-Wright 2005). The stated preference literature has shown that people in general have
a positive attitude towards wind turbines (Borchers et al. 2007), while they at the same time
are able to put a value on the negative externalities related to noise and visual pollution
(Ladenburg 2009; Meyerhoff et al. 2010; Ladenburg and Möller 2011). The stated preference
results are compelling, but a number of questions follow in their wake. For example, when
respondents have to relate to a hypothetical scenario, are they cognitively able to distinguish
between their opinions on noise and visual pollution? If not, are conclusions based on
hypothetical payments as reliable as results based on observed, actual payments? (Diamond
and Hausman 1994).
The externalities related to wind turbines are restricted to local residents, which makes the
hedonic house price method the obvious valuation technique to choose. Only a handful of
hedonic studies have attempted to estimate the local negative impacts of wind turbines and
only the most recent publications have succeeded (Sims and Dent 2007; Sims et al. 2008;
Hoen et al. 2011; Heintzelman and Tuttle 2012). Heintzelman and Tuttle (2012) find that
nearby wind facilities significantly reduce property values. Their results show that property
prices are reduced by between 8.8 % and 14.87 % at a distance of 0.5 miles to the nearest
turbine. They use proximity to wind turbines as a proxy for noise and visual pollution. While
both noise and visual pollution from wind turbines is correlated with proximity, they have a
dissimilar impact and spatial extent. As such, proximity seems to be a rough generalization of
the externalities related to wind turbines, which implies that the result of Heintzelman and
Tuttle (2012) should be interpreted with caution.
While only two hedonic studies have demonstrated that wind turbines have an impact – this
study included – hedonic house price valuation has been used with success on numerous
other externalities, e.g. noise pollution from traffic, having a nice view of, or access to, green
spaces (Day et al. 2007; Sander and Polasky 2009; Zhou et al. 2013). The hedonic literature
on road traffic has treated the related externalities much the same way as wind turbines have
been treated in this study, by explicitly controlling for both view and noise in the hedonic
model. Two examples are Lake et al. (1998) and Bateman et al. (2001). By working with
Geographical Information Systems (GIS), the authors were able to estimate noise and view
variables for each house in their sample. Their conclusions are broadly similar in that noise
and visual pollution from larger roads are reflected in property prices as two different
negative impacts.
The main contribution of this study is the provision of separate estimates of both the noise
and visual pollution from wind turbines. We construct viewsheds based on a high resolution
Digital Surface Model (DSM), which enables us to identify properties where wind turbines
are visible. Noise pollution is calculated for each wind turbine based on noise level
measurements emitted at hub height, distance to the wind turbine, landscape-properties and
2
air absorption under optimal conditions. In total, 12,640 transactions of house sales are
included in the model, which ensured a reasonable variation in the variables of interest.
2. Methods
2.1 Modeling visual pollution
Visual pollution from wind turbines is a unified effect, but it can be subdivided into several
negative effects with different causes, spatial extents and perhaps impacts (Hoen et al. 2011).
Wind turbines in the open landscape can make the area appear more developed and less rural
or less authentic. The general perception of an area can be degraded as can a location with a
scenic view. In addition, wind turbines add movement to the landscape which attracts
attention and reduces the experience of tranquility and peacefulness, which would otherwise
be gained from a rural landscape. The rotating wings of a wind turbine reflect the sun
creating flickers of light, which again attracts attention and adds to the nuisance from the
movement effect. The last visual effect is shadow-flicker. When the wings rotate, they cast a
moving shadow, which in turn causes flickers of shadow in the immediate surroundings of
the wind turbines.
In order to experience a visual effect caused by turbines, one needs to be able to see at least a
part of a turbine. Properties with a view of one or more turbines were identified by
constructing viewsheds for each of the wind turbines in the survey areas at hub height. The
viewshed was based on a high resolution Digital Surface Model (DSM) consisting of 1.6 x1.6
meter cell. The DSM accounts for terrain and obstacles such as buildings, vegetation, forests
and so forth. Houses were identified as having a view of a turbine if at least one of the
corners of the building two meters above terrain was located within the estimated viewshed
of a wind turbine. In total, 33% of the houses in the analysis had a view of a wind turbine.
We captured visual pollution in our model by a dummy variable which indicates whether a
turbine can be seen from the property and by an interaction term between the dummy variable
and the distance to the nearest wind turbine. The specification implies that having a view of a
turbine provides a negative impact and that the impact decreases as distance to the turbine
increases. We assume that the combined negative externalities of the visual pollution of wind
turbines are captured by this specification.
2.2 Noise pollution
Noise from wind turbines stems from three sources; when the wings pass the tower, when the
wings cut through the air and from the mechanics of the turbine. Noise emitted from a turbine
is not constant. Some of the noise is tonal and some is low frequency (Møller et al. 2010).
The composition of the noise affects how the sound is experienced which is different to how
constant noise sources, such as noise from highways, are experienced.
The noise level emissions were calculated for each wind turbine based on how much noise a
turbine emits in the case of optimal conditions for noise production and noise travel distance.
Noise was calculated based on equation (1) which is provided by the Danish legislation in
statute on noise from turbines (Environmental Protection Agency 2011). The equation
describes the noise level emitted from a wind turbine at a given distance measured in decibels
(dB).
3
awaLdBdBhlLSPLΔ−+−+−=5,111)log(*10 22 (1)
Where waL is the sound pressure from the wind turbine provide by the Windpro database
(EMD International A/S 2012), l is the distance to the turbine, h is the hub-height, the 11dB
is a distance correction constant, 1.5 dB is a terrain correction constant assuming a rural
landscape. The air absorption, alΔ , is calculated at by the following equation:
)(1000
2 22hlla+=Δ (2)
Noise levels were divided into noise zones (Table 1). Properties located within these noise
zones were identified by simple overlay analysis in GIS. No house was found to be located
within a noise zone above 50 dB and the majority of houses in the survey area were located
within the noise zone 20-29 dB. Note that a level below 20 dB is generally perceived as
silence (Pedersen and Waya 2004).
TABLE 1 THE DISTRIBUTION OF OBSERVATIONS ACROSS NOISE GROUPS
Noise level <20
dB
20-29
dB
30-39
dB
40-50
dB
Affected properties
(%)
4077
(32)
7532
(60)
879
(7)
152
(1)
Equation (1) does not account for tonal or low frequency noise, which may affect the
perception of experienced noise. Furthermore it does not account for the multiplication effect
of noise-exposure to several wind turbines. Two turbines emit more noise than one. If a house
was affected by more than one wind turbine the house was assigned the highest noise
calculation. In addition the perception of noise may depend on the background noise. The
experience of noise emitted from a turbine in a quiet environment is likely to be perceived
differently from a noisy environment with other external noise sources such as highways or
railways. The noise calculation does not include other sources of noise. However, such
negative externalities are accounted for in the hedonic price model (Table 2).
2.3 Theory
The theoretical foundation for the hedonic valuation method stems from Rosen’s (1974)
seminal paper which demonstrated that buyers and sellers of houses in a perfectly
competitive market will reach a market equilibrium guided by the implicit prices of house
characteristics. Rosen argues that household buyers seek to maximize utility by bidding as
little as possible for every single house (defined by its characteristics) while household sellers
seek to maximize capital rent by offering their house for the highest price possible. The
equilibrium price schedule for house characteristics forms where the bid and offer functions
meet. In equilibrium, the price P of any given house, n, can be modelled as a function of a
vector z that includes all K house characteristics, zik, is:
);,...,...,(1 Θ=nKnknnzzzfP , (3)
4
Where Θ is a set of parameters related to the characteristics and specific to the housing
market considered. Note that the characteristics may also include environmental amenities
and disamenities obtained by ownership of the house, which here relates to whether or not the
property is exposed to visual or noise pollution from wind turbines. Assuming weak
separability with respect to the parameters of interest ensures that the marginal rate of
substitution between any two characteristics is independent of the level of all other
characteristics. With that assumption in place, the implicit price of a house characteristic zk is
its market price and is also a measure of its associated Marginal Willingness To Pay (MWTP)
(Palmquist, 1991) .
In optimum, the household MWTP will equate to the household marginal rate of substitution
between the price of the house characteristic zk and a composite numeraire good, comprising
all other goods. Hence, the slope of the hedonic price function for a given house
characteristic zk can be recognized as the MWTP for house characteristic zk.
nk
n
n dz
dPMWTP= (4)
This allows us to calculate the value of a marginal change in the environmental good also
known as the 1st stage of the hedonic model. From a policy perspective, it can be argued that
the value of such a marginal change in amenity values is seldom a crucial piece of
information. The reason is that the hedonic price function only provides information on one
point on the households’ demand function with respect to the environmental good in question
– not the demand schedule for that good which would be the result of undertaking 2nd stage of
the hedonic model. Nevertheless, results from 1st stage models are the most reported results
in the hedonic literature (Palmquist, 2005) . The main problem in reaching the 2nd stage is
to come up with appropriate instruments to handle the inherit endogeneity which arises when
households both choose the amount and the price of the house characteristics.
2.4 The model
The hedonic house price model is estimated in two steps. In the first step, the sales prices are
detrended using a cross-pooled regression model which allows for different prices across
years and municipalities using 2011 as the reference year. The error term of the cross-pooled
regression consists of logged sales prices detrended in time and space. In the second step, the
hedonic price model is estimated using a simple non-spatial OLS model and two explicit
spatial models based on a Generalized Method of Moments (GMM) estimator developed by
Kelejian and Prucha (2010). The spatial models consist of a Spatial Error Model (SEM) and
a spatial autoregressive model with a spatial autoregressive error term (SARAR). The two
step approach is required because spatial models are not able to identify highly correlated
variables, such as the correlation between the interaction term, the municipalities and the year
dummies in equation 5.
The cross-pooled model which corrects for differences in prices over municipalities and years
can be written as follow:
μββββ++++=tymunicipaliyearyeartymunicipaliP*)ln(3210 (5)
Where ln(P) is logged property prices, 1β is a vector of the parameter estimates for the
dummy variables referring to municipalities, 2β is a vector of the parameter estimates over
5
the 11 year period. 3β is a vector of parameter estimates of the interaction terms between the
municipalities and years. μ is the error term of the model which, essentially, consists of the
logged detrended price and an error term consisting of unexplained noise.
The hedonic house price model is estimated using the logged detrended prices supplied by
Equation 5. The full hedonic SARAR model can be written as follows.
εθθθθμρμ+++++=4321*noisedisviewviewZW (6)
uW+=ελε (7)
Where 1θ is a vector of coefficient estimates of the control variables presented in table 2, 2θ is
the coefficient estimate of the dummy variable of having a view, 3θ is the coefficient estimate
of the interaction term between the view and distance to nearest wind turbine, 4θ represents
the coefficient estimates of being within one of the noise zones using <20dB as reference
zone. W is a row standardized N*N spatial weight matrix based on the 10 nearest neighbors.
The terms ρ and λ are the spatial autoregressive coefficients also known as the spatial lag
term and the spatial error term respectively. The hedonic model is estimated using an (non-
spatial) OLS model, where both ρ and λ are assumed to be zero, a spatial error model, where
ρ is assumed to be zero and λ non-zero, and finally as a SARAR, where ρ and λ are
assumed to be non-zero. The objective of the application of the spatial models is to provide
consistent and efficient parameter estimates that are robust to model specifications and
unobserved spatially correlated variables.
The spatial lag term ρ implies that there is a spillover effect between house prices of
neighboring properties. Lesage and Fischer (2008) distinguish between average direct,
indirect and total impacts, depending on whether one looks solely at the estimated coefficient
or accounts for neighboring observations. From Won Kim et al. (2003), the marginal price of
a housing characteristic (total impact) becomes:
1)(−−=WIdz
d
k
k
ρθμ (8)
Where, I is an identity matrix. The direct effect can be interpreted in the same way as a
standard regression coefficient estimate while the indirect effect depends on the defined
neighbors in the spatial weight matrix. The model suggests a marginal change will set off a
ripple effect through the housing market affecting neighbors and their neighbors and so forth.
We believe that the indirect spillover effect represented by the autoregressive lag term ρ can
be interpreted as an information effect. If buyers and sellers are unsure of the appropriate
value of a property given its characteristics, they may infer the appropriate price by looking
at nearby properties with similar characteristics. The information contained in previous
transactions in the same area may also allow the household to form expectations about the
future evolution of the prices in the area. Alternatively the lagged dependent variable is likely
to be a proxy for unobserved characteristics. In either case, the spill-over effect should be
disregarded in the interpretation of the MWTP in hedonic house price models, as it does not
reflect the preference of buyers.
6
3. Data
In total, the analysis contains 12,640 sales of single-family houses sold over a 12 year period
starting from 2000 to 2011. During this period, several turbines were built. Property prices
prior to turbine construction were modeled as if the property was not exposed to any
externality related to turbines.
Data also contain information on the structural characteristics of the property such as number
of rooms, size of the living area, etc. This information was extracted from the Danish
Registry of Buildings and Housing database (Ministry of Housing Urban and Rural Affairs
2012). The registry also contains information on the exact coordinates of the location of each
house. Proximi1ty variables to environmental externalities were calculated for each property
using ArcGIS Desktop 10.1. The proximity measures are proxies for view, accessibility, etc.
To remove possible border problems, all spatial externalities less than 5.5 kilometres from the
border of the survey areas were included in the calculation of spatial variables. Spatial data
were supplied by the “Danish National Survey and Cadastre” from the spatial database
Kort10 (KMS 2001). A summary of the control variables applied in the model is presented in
Table 2.
TABLE 2 OVERVIEW OF CONTROL VARIABLES IN THE MODEL
Structural variables
Number of floors Number of rooms Brick Tile roof Renovation 1970s
Basement size Number of toilets Flat roof Cement roof Renovation 1980s
Size of living area Number of baths Age Fibre Roof Renovation 1990s
Attic space Low basement Detached house Board roof Renovation 2000s
Environmental variables
Forest Coastal line Highway
Lake Urban zone Large road
Data on wind turbines were provided by the Danish Energy Agency (2012) and include the
geocoded location of the wind turbines, hub height, total height and rotor diameter. Noise
data for each wind turbine were supplied by the database from the planning program
WindPro 2.8 which includes reported noise date from the manufactures (EMD International
A/S 2012). The viewshed of each wind turbine was constructed based on a DSM which
consists of 1.6 x 1.6 meter cells. Each cell contains the average height of the surface which is
defined as ground surface including obstacles relevant to the viewshed such as buildings,
fences, forest, etc. A more detailed description of the properties of DST can be found in
Heywood et al. (2006). The DST was supplied by COWI (2009).
4. Survey area
The survey consists of 24 spatially detached sub-survey areas which, combined, cover 647
km2, 20 municipalities and 55,864 houses in Denmark. The sub-survey areas are located in a
rural environment characterized by fields, small villages and towns.
The main criterion for selection of the survey areas was that they have as many transactions
as possible within a primarily 600 meter and secondarily a 2,500 meter radius of the nearest
wind turbine. The selection criterion resulted in a rather dispersed study area as illustrated in
figure 1. The survey areas were identified using GIS and assessed manually using high
7
resolution aerial photos. Each survey area has a radius of 2.5 kilometres to the nearest wind
turbine, which ensures that the exposure to the wind turbine externality varies between being
exposed to non-exposed.
Figure 1: Map of Denmark showing the spatial distribution of study areas
5. Results
The results of the model estimations are presented in table 3 for wind turbine externalities
and relevant model tests. Model estimates, which include all control variables, are provided
in appendix B. The estimates of wind turbine externalities vary only marginally between
models and are significant at the 5% level except for the view variable in the SEM model and
the 39-50 dB noise zone in the OLS model, which are both significant at the 10 % level. All
three models are robust to heteroscedasticity. The non-spatial model is estimated using OLS
with heteroscedasticity-consistent standard errors. The two spatial models are estimated using
the Generalized Method of Moments estimator (GMM) with innovations robust to
heteroscedasticity, see e.g. Piras (2010) for an elaboration.
Having a view of a wind turbine from your house results in a considerable reduction in the
price schedule of the house. The effect of the view of a wind turbine decreases as distance to
the turbine increases. The models predict that a house located within one of the noise zones
has a discrete impact on the sales price. The negative impact of the noise zone is positively
related to the noise level.
The spatial autoregressive terms in the SEM model and the SARAR model are highly
significant, which indicates that the two models adjust for spatial autocorrelation. Adjusted
R2 is calculated for the three models. The SARAR model has a considerable higher adjusted
R2 than either of the other models. This indicates that the lag term in the SARAR model
improves model performance.
8
Global Moran’s I value is calculated for the residuals for each of the models based on a row
standardized spatial weight matrix, which includes the 10 nearest neighbors. The global
Moran’s I test indicates that all three models suffer from spatial autocorrelation, as the
residuals have a significant spatial structure which is different from a random spatial
distribution.
Spatial dependence of the residuals of the OLS models was tested using Langragian
multiplier statistics. The term robust in the LM-error and LM-lag (in Table 3) indicates that it
tests for one type of dependence under the assumption that the other is present (Anselin et al.
1996). The Langragian multiplier tests are significant for both an error term and lag term. The
error term is the most important of the two terms. In the SARAR model, both autoregressive
terms are included.
TABLE 3 MODEL ESTIMATION OF TURBINE EXTERNALITIES
Variable OLS SEM SARAR
View -0.0168
(0.0134)
*** -0.0315
(0.0172)
. -0.0397
(0.0154)
**
View*distance 6.99E-05
(8.42E-06)
*** 2.42E-05
(1.06E-05)
* 2.78-05
(9.55E-06)
**
20-30dB -0.0368
(0.0056)
*** -0.0306
(0.0101)
** -0.0255
(0.0079)
**
30-39dB -0.0512
(0.0129)
*** -0.0551
(0.0189)
** -0.0442
(0.0151)
**
39-50dB -0.0432
(0.0256)
. -0.06692
(0.0272)
* -0.0509
(0.0247)
*
λ – error term 0.6004
(0.0195)
*** 0.4412
(0.0253)
***
ρ – lag term 0.2677
(0.0253)
***
WALD statistics
( h1:λ= ρ=0)
1538.4***
Adjusted R2 0.3794 0.3704 0.4492
Global Moran’s I 0.2553*** 0.2776*** 0.1367***
LM-error 4629.275***
LM-lag 3220.362***
Robust LM-error 1468.492***
Robust LM-lag 59.576***
N= 12,640, 12,581 degrees of freedom. Standard errors are indicated under estimates in parentheses.
*** significant at 0,1%, ** significant at 1%, * significant at 5%, . Significant at 10 % a)
9
6. Model interpretation
The marginal implicit price of the hedonic price function is presented in table 3. The price-
functions are all log-linear, thus the marginal changes represent the relative change in house
price. Table 4 contains both a marginal willingness to pay in relative and absolute prices
based on the average sales price in 2011 in the survey areas. The table is based on the
estimates of the SEM model. The lag term in the SARAR model implicates a spill-over effect
which may be an information effect. Such an effect would be inappropriate to account for in
the interpretation of the estimates of the hedonic house price model. Given the ambiguous
interpretation of the lag term in the SARAR model, we choose to present and interpret the
estimates of the SEM model (see also section 2.4).
The noise and visual pollution of wind turbines have a considerable impact on local residents.
The hedonic house price models predict that in severe cases roughly 10 % of the sales price
can be explained by exposure to noise and visual pollution from wind turbines. The impact of
turbine noise on the immediate surroundings can be explained by 6.69 % of the house price in
highly exposed areas. The marginal willingness to pay doubles from the low noise zone 20-
30 dB to the high noise zone of 39-50 dB. The visual pollution of a wind turbine can explain
-3.15 % of the house price. Starting from the base of the wind turbine, the price increases by
0.00242 % for each meter away from the turbine for houses which have a view. The
specification of the hedonic model indicates that having a view of a wind turbine is negative.
However, the negative visual impact of the turbine reduces with distance.
The results are in line with the findings of the only other successful hedonic article on the
negative impact of wind turbines. Heintzelman and Tuttle (2012) find a depression in
property price between 8.80% and 14.49% within a radius of 0.5 miles to the nearest turbine.
Our results indicate that prices drop by between 10.31% and 13.94% under similar
circumstances depending on the level of noise exposure.
TABLE 4. MARGINAL IMPLICIT WILLINGNESS TO PAY ESTIMATES
Parameter % of the house price Average MWTP (EUR)
View (dummy) -3.15 % -6,233
View*distance (per 100 meter) 0.242% 479
20-29 dB (dummy) -3.06% -6,055
30-39 dB (dummy) -5.51 % -10,904
40-50 dB (dummy) -6.69% -13,239
7. Conclusion
In this paper we succeeded in separating and identifying the visual and audible externalities
arising from wind turbines. We identified a negative price premium of around 3.15% of the
sales price for having a view of at least one wind turbine. The price premium declines as
distance to the turbine increases at a rate of 0.242% of the sales price per 100 meters.
Furthermore, we find that noise provides an additional negative price premium, which in
terms of impact mirrors that of having a view. Approximately 3% to 7% of the sales price can
be explained by the exposure to noise.
The effect of view and distance changes to a positive externality at around 1,300 meters. Use
of a specification of the turbine distance which allowed for a non-constant effect may have
10
been more appropriate. On the other hand, the variable might capture the duality of the wind
turbine in the rural landscape. While a turbine in the immediate surroundings is experienced
as a negative externality, a wind turbine on the horizon may be a reminder of sustainable
energy production and be perceived positively. The public in general is positive towards
wind-energy although they dislike having it as a neighbor (Devine-Wright 2005). As such,
the specification may capture both the ‘not in my back yard’ aspect and the general positive
attitude towards wind energy. Another interpretation might simply be that properties with a
view of a turbine which is 1,300 meters or more away have a view - which in itself is positive.
The results presented in this article can be applied in cost-benefit analysis especially because
we succeed in modeling view and noise as two separate parameters. Note that the results of
the hedonic house price model only represent marginal willingness to pay and that such
results will not usually be used in scenarios with non-marginal changes. Still, Bartik (1988)
argues that the estimates of non-marginal localized changes based on the hedonic house price
model can be used as estimates of benefits or costs, given that the non-marginal change is
restricted to a local area, thus not affecting the global housing market. We regard setting up a
wind turbine in the landscape to be both non-marginal and localized. Based on this
assumption, our results are directly applicable in the planning process and could be used to
compensate those living close to wind turbines, or as part of a welfare economic cost-benefit
analysis that includes the negative effects of noise and visual pollution.
In the analyses we do not account for a possible accumulation effect of wind turbines. The
effect of having one wind turbine as opposed to having several turbines or an entire wind
farm may be very different. We only account for the nearest turbine and disregard a possible
accumulation effect. In addition, information on manufacture and turbine production capacity
has been ignored. Such information might have provided further relevant results.
The analysis covers a large number of spatially detached areas. Recall that the hedonic price
schedule is assumed to be generated in an equilibrium market. We essentially assume that the
supply and preference structure are stable across the spatially detached areas and recognize
that this might not be a fully valid assumption. Parameter estimates of noise and view
between municipalities in the survey areas were tested by an ANOVA test. Based on this, we
cannot reject that parameter estimates between municipalities are different. Previous hedonic
studies on wind turbines have very likely suffered from lack of spatial variation due to a
small dataset (Heintzelman and Tuttle 2012). The number of survey areas chosen in this
analysis ensures a reasonable variation in the wind turbine variables.
Neither of the model estimations fully resolves the problem of spatial autocorrelation. Both
explicit spatial models retain a significant spatial structure in the error term. This indicates
that the models still suffer from omitted spatial processes such as misspecification of the
functional form, mis-measurement of spatial covariates or from omitted spatial covariates. If
the omitted spatial processes are not correlated with the turbine variables, the estimate of the
impact of wind turbines remains trustworthy. In addition, the model estimates are robust
across models.
We conclude that noise and visual pollution from wind turbines have a considerable impact
on nearby residential properties. When Don Quijote was tilting at windmills he was fighting
imaginary giants. At present, wind turbines are a symbol of sustainable energy, the way of
the future. However, local residents who live in close proximity to these sustainable giants
experience some very real negative externalities in the form of noise and visual pollution.
11
8. Literature
Bateman, Ian, Brett Day, Iain R. Lake, Andrew A. Lovett, 2001. The Effect of Road Traffic
on Residential Property Values: A Literature Review and Hedonic Pricing Study.
Development Department, Edinburgh, School of Environmental Sciences, University
of East Anglia, Norwich
Borchers, Allison M., Joshua M. Duke, George R. Parsons. 2007. "Does willingness to pay
for green energy differ by source?" Energy Policy 35, 3327-3334.
COWI, 2009. Digital Surface Model.
Danish Energy Agency, 2012. Stamdataregister for vindmøller
Day, Brett, Ian Bateman, Iain Lake. 2007. "Beyond implicit prices: recovering theoretically
consistent and transferable values for noise avoidance from a hedonic property price
model" Environmental and Resource Economics 37, 211-232.
Devine-Wright, Patrick. 2005. "Beyond NIMBYism: towards an integrated framework for
understanding public perceptions of wind energy" Wind Energy 8, 125-139.
Diamond, Peter A., Jerry A. Hausman. 1994. "Contingent Valuation: Is Some Number Better
Than No Number" Journal of economic Perspective 8, 45-64.
EMD International A/S, 2012. Wind Pro.
Environmental Protection Agency, 2011. Bekendtgørelse om støj fra vindmøller (Statue on
noise from turbines).
Heintzelman, Martin D., Carrie M. Tuttle. 2012. "Values in the Wind: A Hedonic Analysis of
Wind Power Facilities" Land Economics 88, 547-588.
Heywood, Ian, Sarah Cornelius, Steve Carver, 2006. An introduction to geographical
information systems.3th Harlow England.Education Limited.
Hoen, Ben, Ryan Wiser, Peter Cappers, Mark Thayer, Gautam Sethi. 2011. " Wind Energy
Facilities and Residential Properties: The Effect of Proximity and View on Sales
Prices" Journal of Real Estate Research 3, 279-306.
Kelejian, Harry H., Ingmar R. Prucha. 2010. "Spatial models with spatially lagged dependent
variables and incomplete data" Journal of Geographical Systems 12, 241-257.
KMS, 2001. TOP10DK Geometrisk registrering Specifikation udgave 3.2.0.
Ladenburg, Jacob. 2009. "Stated public preferences for on-land and offshore wind power
generation-a review" Wind Energy 12, 171-181.
Ladenburg, Jacob, Bernd Möller. 2011. "Attitude and acceptance of offshore wind farms—
The influence of travel time and wind farm attributes" Renewable and Sustainable
Energy Reviews 15, 4223-4235.
Lake, Iain R., Andrew A. Lovett, Ian J. Bateman, Ian H. Langford. 1998. "Modelling
environmental influences on property prices in an urban environment" Computers
Environment and Urban Systems 22, 121-136.
Lesage, James P., Manfred M. Fischer. 2008. "Spatial Growth Regressions: Model
Specification, Estimation and Interpretation" Spatial Economic Analysis 3, 275-304.
Meyerhoff, Jürgen, Cornelia Ohl, Volkmar Hartje. 2010. "Landscape externalities from
onshore wind power" Energy Policy 38, 82-92.
Ministry of Housing Urban and Rural Affairs, 2012. Den Offentlige Informationsserver.
Ministry of Housing Urban and Rural Affairs, http://www.boligejer.dk/om-
ejendomsdata: Grontmij
Møller, H., A. Pedersen, J. Staunstrup, 2010. Støj fra testcenter for vindmøller Østerild.
Sektion for Akustik Aalborg Universitet, Sektion for Geoinformatik og
Arealforvaltning,
Palmquist, R. , 1991. "Hedonic methods". In Measuring the Demand for Environmental
Quality, ed. John E. Braden and Charles D. Kolstad, Amsterdam: Elsevier Science
12
Palmquist, Raymond B., 2005. "Property Value Models". In Handbook of Environmental
Economics, ed. Karl-Göran Mäler and Jeffrey R. Vincent, Amsterdam: Elsevier.
Pedersen, Eja, Kerstin P. Waya. 2004. "Perception and annoyance due to wind turbine
noise—a dose–response relationship" Acoustical Society of America, 3460–3470.
Piras, Gianfranco. 2010. "Sphet: Spatial Models with Heteroskedastic Innovations in R"
Journal of Statistical Software 35, 1-21.
Rosen, Sherwin. 1974. "Hedonic Prices and Implicit Markets: Product Differentiation in Pure
Competition" The Journal of Political Economy 82, 34-55.
Sander, Heather A., Stephen Polasky. 2009. "The value of views and open space: Estimates
from a hedonic pricing model for Ramsey County, Minnesota, USA" Land Use Policy
26, 837-845.
Sims, Sally , Peter Dent. 2007. " Property stigma: wind farms are just the latest fashion"
Journal of Property Investment and Finance 25, 626-651.
Sims, Sally, Peter Dent, Reza O. Oskrochi. 2008. "Modelling the impact of wind farms on
house prices in the UK" International Journal of Strategic Property Management 12,
251-269.
Wolsink, Maarten. 2000. "Wind power and the NIMBY-myth: institutional capacity and the
limited significance of public support" Renewable Energy 21, 49-64.
Won Kim, Chong, Tim T. Phipps, Luc. Anselin. 2003. "Measuring the benefits of air quality
improvement: a spatial hedonic approach" Journal of Environmental Economics and
Management 45, 24-39.
Zhou, Qianqian, Toke E. Panduro, Bo J. Thorsen, Karsten Arnbjerg-Nielsen. 2013.
"Adaption to Extreme Rainfall with Open Urban Drainage System: An Integrated
Hydrological Cost-Benefit Analysis" Environmental Management, 1-16.
13
Appendix A.1 Descriptive statistics
TABLE A1 DESCRIPTIVE STATISTICS FOR DUMMY VARIABLES
Name Description Mean Observations =1
Brick House build in bricks 0.9158 13592
Flat roof Flat roof 0.0244 362
Cement roof Cement roof 0.1979 2937
Fibre roof Fibre roof 0.4445 6597
Board roof Board roof 0.0268 398
Tile roof Tile roof 0.2778 4123
Lower basement Lower basement 0.0912 1354
Detached house The property is a detached house 0.8213 12189
Renovation 1970s House rebuilt between 1970-1979 0.1113 1652
Renovation 1980s House rebuilt between 1980-1989 0.0703 1044
Renovation 1990s House rebuilt between 1990-1999 0.0551 817
Renovation 2000s House rebuilt between 2000-2009 0.0701 1041
<20 dB
Within a zone where a turbine makes noise <20
dB
0.3181 4721
20-30dB Within a zone where a turbine makes noise 20-30
dB
0.5908 8768
30-39dB Within a zone where a turbine makes noise 30-
39dB
0.0756 1122
39-50dB Within a zone where a turbine makes noise 39-
50dB
0.0151 224
View At least one turbine is visible 0.3547 5264
Urban zone House within urban zone or not 0.8117 12047
14
A.2 DESCRIPTIVE STATISTICS FOR NON-DUMMY VARIABLES
Name Description Mean Min Max
Price Trade price (not corrected for
inflation) in KKR 1329000 100000 18150000
Age Age of the house 1957 1850 2010
Number of baths Number of bathrooms 1.268 1 4
Size Size of living area m2 136.5 56 492
Basement size Size of basement m2 12.19 0 230
Attic size Size of attic m2 24.43 0 260
Number of rooms Number of rooms 4.642 1 16
Number of floors Number of floors 1.03 1 3
Number of toilets Number of toilets 1.536 1 5
Number of bathrooms Number of bathrooms 1.268 1 4
Forest
Distance in meters to the nearest
forest, zero being within forest.
In the model used as dummy
variables based on steps of 100
meters with reference distance
being above 700 meters. 297.2 0 4294
Lake
Distance in meters to the nearest
lake with a surface greater than
200m2. In the model used as
dummy variables based on steps
of 100 meters with reference
distance being above 700 meters. 4390 0 10500 (1903)
Coast line
Distance in meters to the nearest
coastal line. In the model used as
dummy variables based on steps
of 100 meters with reference
distance being above 700 meters. 4677 8.022 10500 (2348)
Highway
Distance in meters to the nearest
highway. In the model used as
dummy variables based on steps
of 100 meters with reference
distance being above 1000
meters. 8297 17.94 10500 (9606)
Large Road
Distance in meters to the nearest
road wider than 6m. In the model
used as dummy variables based
on steps of 100 meters with
reference distance being above
400 meters. 393.9 2.847 5239
Distance
Distance in meters to the nearest
onshore turbine 1489 68.27 3474
15
Appendix B Model estimation
TABLE B.1 MODEL ESTIMATION
Variable OLS SEM SARAR
Intercept
-4.898 ***
(0.1764)
-5.5614 ***
(0.2496)
-5.1706 ***
(0.2193)
Brick
0.06860 ***
(0.009516)
0.0625 ***
(0.0102)
0.0603 ***
(0.0096)
Tile roof
0.01563
(0.01681)
-0.0069
(0.0161)
-0.0042
(0.0155)
Cement roof
-0.02654
(0.01725)
-0.0300.
(0.0167)
-0.0295 .
(0.0161)
Fibre roof
-0.1038 ***
(0.01634)
-0.0933 ***
(0.0152)
-0.0295 .
(0.0161)
Board roof
-0.03898 .
(0.02218)
0.0617 **
(0.0225)
-0.0552 *
(0.0215)
Flat roof
-0.1393 ***
(0.02205)
-0.1312 ***
(0.0208)
-0.0552 *
(0.0215)
Age
0.001024 **
(0.00009)
0.0014 ***
(0.0001)
0.0012 ***
(0.0001)
Detached house
0.01787 *
(0.007839)
0.0227 *
(0.0091)
0.0080
(0.0082)
Number of bathrooms
0.05391 ***
(0.005847)
0.0432 ***
(0.0060)
0.0420 ***
(0.0058)
Low basement
0.01800 *
(0.008778)
0.0228 *
(0.0091)
0.0264 **
(0.0087)
Size (log)
0.5547 ***
(0.01042)
0.5386 ***
(0.0109)
0.5333 ***
(0.0105)
Basement Size
0.0006684 ***
(0.00009)
0.0007 ***
(0.0001)
0.0008 ***
(0.0001)
Renovation 1970s
-0.03566 ***
(0.008036)
-0.0261 ***
(0.0069)
-0.0247 ***
(0.0067)
Renovation 1980s
0.006676
(0.009700)
0.0069
(0.0087)
0.0061
(0.0085)
Renovation 1990s
0.09875 ***
(0.01092)
0.0995 ***
(0.0101)
0.1032 ***
(0.0099)
Renovation 2000s
-0.09579 ***
(0.01018)
-0.0930 ***
(0.0113)
-0.0924 ***
(0.0110)
Urban zone
0.005843
(0.008138)
0.0275.
(0.0163)
0.0173
(0.0125)
Coast 0-100 meters
0.2965 ***
(0.03116)
0.3607 ***
(0.0477)
0.2787 ***
(0.0407)
Coast 101-200 meters
0.1765 ***
(0.02071)
0.2316 ***
(0.0355)
0.1613 ***
(0.0284)
Coast 201-300 meters
0.1685 ***
(0.01724)
0.2309 ***
(0.0328)
0.1739 ***
(0.025)
Coast 301-400 meters
0.1549 ***
(0.01627)
0.1587 ***
(0.0307)
0.1262 ***
(0.0245)
Coast 401-500 meters
0.1783 ***
(0.01588)
0.1427 ***
(0.0296)
0.1275 ***
(0.0236)
Coast 501-600 meters
0.1090 ***
(0.01594)
0.1023 ***
(0.0302)
0.0914 ***
(0.0235)
Coast 601-700 meters
0.07233 ***
(0.01.818)
0.0549 *
(0.0268)
0.0488 *
(0.0227)
Highway 0-100 meters
-0.4143 **
(0.1401)
-0.3882 ***
(0.1032)
-0.3927 ***
(0.1077)
16
Variable OLS SEM SARAR
Highway 101-200 meters
-0.2169 *
(0.08812)
-0.1587.
(0.0930)
-0.1388 .
( 0.0773)
Highway 201-300 meters
0.1733
(0.1126)
0.1137.
(0.0690)
0.1196 *
(0.0567)
Highway 301-400 meters
-0.001535
(0.08718)
-0.0564
(0.0599)
-0.0629
(0.0538)
Highway 401-500 meters
0.1871 ***
(0.04459)
0.1503.
(0.0817)
0.1403 *
(0.0615)
Highway 501-600 meters
0.1545 **
(0.04750)
0.1423 **
(0.0451)
0.1240 **
(0.0378)
Highway 601-700 meters
0.08173 **
(0.02806)
0.0724
(0.0449)
0.0542
(0.0344)
Highway 701-800 meters
0.1393 ***
0.03007)
0.1283 ***
(0.0342)
0.1080 ***
(0.0267)
Highway 801-900 meters
0.09568 **
(0.03280)
0.1021 **
(0.0328)
0.0797 **
(0.0265)
Highway 901-1000 meters
0.1141 **
(0.03714)
0.0561
(0.0378)
0.0505
(0.0323)
Forest 0-100 meters
0.1012 ***
(0.01335)
0.0953 ***
(0.0259)
0.0605 **
(0.0198)
Forest 101-200 meters
0.08487 ***
(0.01322)
0.0806 **
(0.0256)
0.0491 *
(0.0196)
Forest 201-300 meters
0.08442 ***
(0.01327)
0.0826 **
(0.0257)
0.0531 **
(0.0196)
Forest 301-400 meters
0.09466 ***
(0.01366)
0.0995 ***
(0.0262)
0.0679 ***
(0.0199)
Forest 401-500 meters
0.1067 ***
(0.01467)
0.1010 ***
(0.0271)
0.0716 ***
(0.0208)
Forest 501-600 meters
0.1152 ***
(0.01666)
0.0930 **
(0.0284)
0.0709 **
(0.0222)
Forest 601-700 meters
0.09010 ***
(0.01902)
0.0712 *
(0.0287)
0.0604 *
(0.0236)
Lake 0-100 meters
0.3625 ***
(0.03402)
0.3605 ***
(0.0591)
0.2606 ***
(0.0494)
Lake 101-200 meters
0.2023 ***
(0.02188)
0.1990 ***
(0.0406)
0.1200 ***
(0.0322)
Lake 201-300 meters
0.07006 ***
(0.01948)
0.0930 **
(0.0334)
0.0368
(0.0270)
Lake 301-400 meters
0.03122
(0.01909)
0.0528 *
(0.0265)
0.0241
(0.0220)
Lake 401-500 meters
-0.03926 *
(0.01567)
-0.0326
(0.0239)
-0.0434 *
(0.0186)
Lake 501-600 meters
-0.006999
(0.01693)
0.0078
(0.0230)
-0.0085
(0.0182)
Lake 601-700 meters
0.008090
(0.01672)
0.0250
(0.0215)
0.0108
(0.0176)
Large road 0-100 meters
-0.0007251
(0.07053)
-0.0120
(0.0130)
-0.0042
(0.0097)
Large road 101-200 meters
0.03311 ***
(0.007329)
0.0182
(0.0126)
0.0260 **
(0.0096)
Large road 201-300 meters
0.02528 **
(0.007705)
-0.0007
(0.0127)
0.0075
(0.0099)
Large road 301-400 meters
0.02135 *
(0.08466)
0.0034
(0.0116)
0.0084
(0.0095)
View -0.1173 ***
(0.01340)
-0.0288.
(0.0172)
-0.0373 *
(0.0153)
View*distance 7.018e-05 *** 2.2500E-05 * 2.6303E-05 *
17
Variable OLS SEM SARAR
(8.426e-06) (1.0700E-05) (9.5176E-06)
20-30dB -0.03674 ***
(0.005936)
0.0290 **
(0.0106)
-0.0253 **
(0.0082)
30-40dB -0.05112 ***
(0.01180)
-0.0551 **
(0.0194)
-0.0462 **
(0.0153)
40-50dB -0.04604 .
(0.02444)
-0.0642*
(0.0279)
-0.0512 *
(0.0247)
SPATIAL ERROR TERM (ρ) 0.6280 ***
(0.0123)
0.4742 ***
(0.0249)
SPATIAL LAG TERM (λ) 0.2617 ***
(0.0282)
WALD statistics ( h1:λ= ρ=0) 1556,6 ***
R2 0.3823 0.3724 0.4552
N= 12640, 12581 degrees of freedom
*** significant at 0,1%, ** significant at 1%, * significant at 5%, . Significant at 10 %
(): Standard error, R2 for the OLS adjusted, for SEM and GSM pseudo-R2