Ecological Economics 111 (2015) 1222

Contents lists available at ScienceDirect

Ecological Economics
journal homepage: www.elsevier.com/locate/ecolecon

Analysis

What inuences the probability of wind farm planning approval:

Evidence from Ireland
Thomas M. van Rensburg a, Hugh Kelley a,, Nadine Jeserich a,b
a
b

Department of Economics, Cairnes School of Business and Economics, National University of Ireland, Galway, Galway, Ireland
GrowthEconomics, Inc. 2425 Gulf of Mexico Dr., Longboat Key, FL 34228, USA

a r t i c l e

i n f o

Article history:
Received 11 February 2014
Received in revised form 27 October 2014
Accepted 27 December 2014
Available online 22 January 2015
Keywords:
Wind-energy
Planning-approval
Revealed-preference

a b s t r a c t
The purpose of this work is to explore the extent to which wind farm planning approvals in the Republic of
Ireland are inuenced by project technology, institutional processes, and site endowments. We use principal
components data reduction, z-score data normalization, and Probit regression analyses on a unique revealed
preference dataset covering 354 wind farm applications and planning authority decisions between 1990 and
2011. Notably, a unique measure of variable importance is employed that mitigates statistical problems and allows for the ranking of predictors according to their relative inuences. Findings reveal that the duration of the
local appeal process, decisions of local authorities and inspectors, identities of the appellants, and, projects that
conict with strategic development plans or generate visual externalities emerge as key inuences affecting
planning approval. Project technology features such as area, rated output capacity, and hub height, as well as
site wind endowments, appear to be of less but signicant importance. Alternatively, we nd that proximity to
dwellings, towns, or protected habitats does not inuence planning outcomes.
2015 Elsevier B.V. All rights reserved.

1. Introduction
An important question within the wind energy literature concerns
the factors that are empirically associated with wind farm (WF) planning approvals. Having reliable predictors of project planning approval
can assist planners and investors to develop better applications, and
can help all parties better assess the risks and returns of WF projects.
Unfortunately, only a few studies have attempted to rigorously quantify
key inuences on historical planning outcomes (Haggett and Toke,
2006; Horst, van der and Toke, 2010; Toke, 2005a, 2005b). Most other
studies rely upon ndings from stated preference surveys and choice
experiments investigating public attitudes and perceptions about
wind power (Alvarez-Farizo and Hanley, 2002; Bergmann et al., 2006;
Longo et al., 2008; Yadav et al., 2012). Such studies nd conicting positive (Eltham et al., 2008) and negative attitudes (Devine-Wright,
2005a; Landenburg and Dubgaard, 2007; Meyerhoff et al., 2010;
Navrud and Braten, 2007), where the latter responses are more common the greater the degree of proximity of the project to the respondents' residences, i.e. NIMBYism. Although such stated preference
studies are informative, for the most part they are hypothetical ex
ante studies focusing on public preferences rather than outcomes and
therefore they do not yield many practical insights for those involved
in the planning approval process. This limitation could lead to
Corresponding author.
E-mail addresses: thomas.vanrensburg@nuigalway.ie (T.M. van Rensburg),
hugh.kelley@nuigalway.ie (H. Kelley), nadine@growtheconomics.com (N. Jeserich).

http://dx.doi.org/10.1016/j.ecolecon.2014.12.012
0921-8009/ 2015 Elsevier B.V. All rights reserved.

circumstances where the decisions of planning authorities appear random or inuenced by unobserved factors, potentially leading applicants
to perceive political as well as nancial risk when evaluating projects.
This may cause risk averse developers to refrain from initiating projects,
and this in turn could compromise a nation's ability to enhance the
share of renewables in their energy generation portfolios and to comply
with government directives. See the European Directive on Electricity
Production from Renewable Energy Sources (Mguez et al., 2006), and
the U.S. Renewable Electricity Standard (Delmas and Montes-Sancho,
2011).
Despite the limitations of some of these studies, one of their contributions has been the listing and categorization of potential inuences
on the likelihood of WF success. Based on this work we propose that
key predictors of planning approval success generally relate to project
technology, institutional processes, and site endowment characteristics;
several control variables are also included. By analyzing all these variables in an integrated framework, and by normalizing our non-binary
data, we can identify signicant inuences, we can compare the relative
magnitudes of these associations, and we can address key questions in
the literature. Our method follows from Kelley (1998), Jeserich et al.
(2012), Laepple and Kelley (2013), and Kelley et al. (under review),
and allows us to mitigate a variety of statistical problems that plague
this data. Several robustness checks conrm that our approach addresses these issues and provides robust parameter estimates.
There are multiple objectives of this study. A methodological objective is to correct for a variety of statistical problems within our revealed
preference dataset, including variations in predictor distributional

T.M. van Rensburg et al. / Ecological Economics 111 (2015) 1222

properties and magnitudes, the presence of outliers, and predictor collinearity. We do this by sequentially employing z-score data normalization, principal components analysis (PCA) data reduction, and then
Probit analyses. Our quantitative objective is to address two research
questions. First, what are the relative magnitudes of the associations
among wind farm project planning approvals and control, institutional
process, wind farm project, and location endowment attributes? Second, how can controllable (by an involved party) attributes be used to
inuence the probability of approval?

13

in between these extremes). The remaining four categories describe

control, endowment, process, and project predictor variables. For the
purpose of avoiding the dummy variable trap, predictor variables are
further disaggregated into topical sub-categories and one variable
from each sub-category is dropped so as to facilitate intuitive interpretation of remaining parameters. The sub-categories include institutional
process, ecological effects, human welfare effects, and project technology sub-categories.
2.1. Study Area

2. Data and Variables

The datasets for this study were laboriously assembled from public
data sources including Irish county level planning authorities, the national appeals board An Bord Pleanala, the Sustainable Energy Authority
of Ireland wind speed database, and from company records for applicants. This revealed preference data describes actual planning outcomes
and project characteristics, and are grouped into ve categories. The
rst includes the multiple variables reecting different levels of the application process that are used to construct our planning approval dependent variable. This variable captures planning approval success
along any of the multiple paths (e.g. immediate local approval with no
objections, approval only after all appeal attempts, and all variations

The physical setting within which we investigate WF planning approval is the Republic of Ireland between 1990 and 2011. Our data set
covers 354 projects, out of an estimated total of 1300, for 23 of the 26
counties in the Republic of Ireland. These include all counties except,
for obvious reasons, the urban counties of Dublin, Kildare and Meath.
Table 1 provides descriptive statistics for this data and Table A.1 in
Appendix A provides detailed descriptions of our variables and provides
an intuitive sense about what they describe. For the binary data in
Table 1, i.e. variables with a minimum of 0 and a maximum of 1, a
value of 1 indicates that the described variable name/outcome occurred,
unless otherwise noted. For example, the 1st Inspector Granted Approval equal to one indicates this happened. The units for the continuous

Table 1
Descriptive statistics for wind farm data for Republic of Ireland between 1990 and 2011. Number of observations N = 354.
Variable

Variable type

Mean

S.D.

Min.

Max.

Project approval
Closest town (km)
WindEndowment (ave mps)
Appeal1Duration (days)
Time2Ap1 (days)
TotLocDuration (days)
1st Inspect Grant (1 = yes)
1st Inspect Refuse (1 = yes)
AppealApplicant (1 = yes)
AppealGroup (1 = yes)
AppealAffectParty (1 = yes)
AppealThirdParty (1 = yes)
Bord Withdrawn (1 = yes)
LA refused (1 = yes)
REF archeology (1 = yes)
REF Birds (1 = yes)
REF Cum. Impacts (1 = yes)
REF EIA (1 = yes)
REF Flooded area (1 = yes)
REF Flora (1 = yes)
REF Habitat (1 = yes)
REF Peat Stability (1 = yes)
REF Prox Natura (1 = yes)
REF Aviation (1 = yes)
REF StratDevPlan (1 = yes)
REF Prox Dwelling (1 = yes)
REF Public Safety (1 = yes)
REF Tourism (1 = yes)
REF Visual (1 = yes)
ApplicantIndividual (1 = yes)
ApplicantLtd (1 = yes)
ApplicantSubsid. (1 = yes)
ApplicantCountry (1 = ROI)
Contract Amend. (1 = yes)
Contract Exten. (1 = yes)
Perm Extension (1 = yes)
Permission (1 = yes)
Windfarm Exten (1 = yes)
EIS (1 = yes)
Area (ha)
Hub height (m)
No. of turbines (#)
Rated OutCap (MW)
Rotor Diameter (m)

Dep. var
Control
Endowment
Process:Insitut.
Process:Insitut.
Process:Insitut.
Process:Insitut.
Process:Insitut.
Process:Insitut.
Process:Insitut.
Process:Insitut.
Process:Insitut.
Process:Insitut.
Process:Insitut.
Process:Ecol. Effect
Process:Ecol. Effect
Process:Ecol. Effect
Process:Ecol. Effect
Process:Ecol. Effect
Process:Ecol. Effect
Process:Ecol. Effect
Process:Ecol. Effect
Process:Ecol. Effect
Process:Human Effect
Process:Human Effect
Process:Human Effect
Process:Human Effect
Process:Human Effect
Process:Human Effect
Project:General
Project:General
Project:General
Project:General
Project:General
Project:General
Project:General
Project:General
Project:General
Project:General
Project:Technology
Project:Technology
Project:Technology
Project:Technology
Project:Technology

0.20
4.30
8.12
66.96
68.12
175.04
0.00
0.19
0.22
0.01
0.02
0.18
0.02
0.28
0.00
0.02
0.01
0.01
0.00
0.01
0.01
0.02
0.01
0.01
0.12
0.05
0.02
0.01
0.17
0.38
0.57
0.08
0.96
0.02
0.02
0.02
0.93
0.01
0.63
37.20
55.67
6.16
10.66
55.89

0.38
2.81
0.70
139.17
334.13
244.39
0.05
0.38
0.41
0.12
0.15
0.38
0.13
0.44
0.05
0.14
0.11
0.09
0.05
0.07
0.09
0.14
0.11
0.07
0.31
0.20
0.12
0.09
0.36
0.48
0.49
0.26
0.20
0.14
0.15
0.14
0.25
0.12
0.42
79.49
19.89
6.84
11.71
23.50

0
0
6.29
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
6
1
0
1.70

1
19.00
10.79
734.12
2000
2000
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1092.00
115.77
48
105.00
114.00

14

T.M. van Rensburg et al. / Ecological Economics 111 (2015) 1222

variables are indicated in parentheses next to the variable name. In general one can see that the projects come from a diverse set of private individual and corporate ownership structures, they represent different
types of planning approval or contract requests, have different inspector
outcomes, appeal types, local and national approval board outcomes,
and different appellant refusal reasons. Importantly, the data display
variation across a diverse set of attributes which indicates there is a
wealth of cross sectional variation to associate with our planning approval dependent variable.
2.2. Dependent Variable: Project Planning Approval
To determine if a project is actually approved we need to account for
the fact that projects follow a variety of paths that may result in the nal
project approval or refusal. Importantly, this work focuses on the planning approval process rather than project build/construction outcome.
This is because approval does not necessarily imply a project is built,
in fact only a subset of approved projects are actually built in ROI.
These approval paths reect the potentially sequential/conditional possibilities of immediate approval, immediate rejection, appeals by both
project initiators and affected local residents, and decisions by both
local authorities and the national appeals board. Accounting for these
paths requires the integration of information from local inspectors and
residents, and local and national levels of government. In other words,
there is no one source for a simple project built or project approved variable. Instead, we must consider if local authority granted (LAG), local
authority granted conditionally (LAGC), if there was a rst appeal date
(AP1D), whether An Bord Pleanala refused (BR), or An Bord Pleanala
granted (BG), or if An Bord Pleanala granted conditionally (BGC). For
simplicity, let the two levels of appeals, local granted (LG) and national
granted (B) be represented as LG = LAG + LAGC and B = BG + BGC.
The nal dependent variable then represents initial or eventual project
approval via any path and is simply,
ProjectApproval LG  B LG AP1D 0
AP1D0  B LG  BRBR 1LG B:
1
In Eq. (1) equalities and inequalities indicate Boolean tests resulting
in a value of one or zero if the condition is true or false respectively. Considering a specic and common example of the path projects take to approval can provide some intuition. Assume that the local authority
grants with conditions (LAGC = 1, LAG = 0, implying LG = 1), the applicant then appeals (presumably the conditions) resulting AP1D = 1 (0
reects the absence of an appeal or appeal date), and nally the national
authorities grant with conditions (i.e. refuse the applicant's appeal)
leading to BGC = 1, BG = 0, and therefore B = 1. These values indicate
that ProjectApproval = 1 [1+0] + 1 [1 (0 1) + 1] = 1. An immediate local refusal, appeal, and then national refusal would result in
LG = 0, AP1D = 1, B = 0, BR = 1, and overall ProjectApproval =
0 [1 + 0] + 1 [1 1+0] = 0. Intuitively, this measure summarizes
all potential paths to planning success, and takes a value of one if the
project is approved by any path, and zero otherwise. The top row of
Table 1 provides summary statistics for this variable, and indicates
that 20% of the projects in our sample are ultimately approved by any
path. The large (0.38) standard deviation for this dependent variable
suggests substantial variation and an excellent opportunity to identify
predictor associations.
2.3. Independent Variables: Control, Endowment, Process, and Project
Our predictor variables are binary and continuous, and span control,
endowment, process, and project categories; variable names are provided within parentheses below. In terms of control data we include binaries for local characteristics such as the county of project residence
(County X), and a continuous variable describing proximity to the

closest town (Closest town). The former is motivated by earlier planning and appeals literature suggesting that outcomes may be strongly
inuenced by local conditions or proximity to local stakeholders
(Devine-Wright, 2005b; Horst, van der and Toke, 2010; Meyerhoff
et al., 2010; Navrud and Braten, 2007). For brevity we do not report
the summary statistics for the county dummy variables; however we
observe that Country 2 (Clare) and Country 1 (Donegal) have the largest
percentages of wind farm applications in the dataset, at 12% and 8% respectively. The remaining counties represent each about 34% of the
sample. The other control variable summary statistics indicate that, on
average, the closest town to a project is 4 km away, and in some cases
they lay within city limits, i.e. the minimum distance of 0, or are as far
away as 19 km. If parameters for county dummy variables are signicant, this would suggest that unique county characteristics inuence
the probability of planning approval. If the parameter on proximity is
signicant, this yields insights about NIMBYism.
Our endowments data is based upon the average of the wind speeds
at various mast heights (wind endowment). The motivation for including this variable is that, all else being equal, wind projects with high revenue potential, which can be proxied by wind endowments or
interactions with it, should have a higher probability of approval (Lee
et al., 2009; Mudasser et al., 2013). This could occur because of community prot sharing benets, because of the high return for the wind project and the tax revenue this produced at the national level, due to
contributions to long-term local business sustainability (Ellis and
SQW, 2012), or because the return is more likely to exceed the opportunity cost of removing land from agricultural production (where most
projects reside). Table 1 indicates that the average multi-height wind
speed is 8.1 meters per second (mps), and it ranges from a minimum
of 6.3 to a maximum of 10.8 mps. This is high compared to the values
for other countries, see Sahu et al. (2013) and Le Gourieres (2014),
who dene high average wind speed sites/countries to be larger than
6.9 mps. This suggests that ROI may have a comparative advantage in
the production of wind energy.
Our Project predictor variables are disaggregated into general and
technology related subcategories. Earlier work suggests that local stakeholder ownership or nancial participation appears as an important
predictor of planning outcomes (Maillebouis, 2003; Devlin, 2005;
Devine-Wright, 2005b; Wolsink, 2006; Toke, 2005a, 2005b; Toke
et al., 2008; Dimitropoulos and Kontoleon, 2009; Warren and
McFadyen, 2010). We therefore include general project information
about whether the project involves individual (ApplicantIndividual),
corporate (ApplicantLtd), and/or subsidiary corporate (Applicant
Subsid.) ownership structures, as well as information about applicants'
nationality (ApplicantCountry = 1 if Irish). We see that 57% of projects
are proposed by limited liability corporations, 38% by individuals, and
8% by a subsidiary corporation. Note that the sum of the three applicant
types, individual, Ltd., and subsidiary does not exactly sum to one because 12 of the 354 projects had both corporate and subsidiary applicant
classications. And, there is a fourth category other which is dropped
to avoid the dummy variable trap. For the former case, this indicates a
subsidiary and a parent company (not necessarily of the subsidiary)
jointly proposed a project. In contrast, Ltd. or subsidiary represents
only a parent company or subsidiary proposing a project. Additionally,
96% of the projects in our sample are proposed by companies or individuals resident in ROI. Another important general project characteristic
likely to affect planning outcomes involves the type of project approval
sought. These types include requests for contract amendments to
existing contracts (Contract Amend.; 2% of total projects), contract extensions (Contract Exten.; 2%), planning permission extensions (Perm
Extension; 2%), requests to extend existing wind farms (Windfarm
Exten.; 1%), or rst time permission requests for new wind farms (Permission; 93%). Finally, whether environmental impact assessments are
conducted can inuence the planning approval outcome (Valentine,
2011). We observe that 63% of the projects in our sample included
these assessments (EIS). Project technology characteristics related to

T.M. van Rensburg et al. / Ecological Economics 111 (2015) 1222

the turbines, the masts, and the scale of the project are also likely to affect the planning process (Devine-Wright, 2005a; Dimitropoulos and
Kontoleon, 2009; Meyerhoff et al., 2010). These features have implications in terms of protability/revenue potential (positive expected association), and in terms of project foot print/negative landscape
externalities (negative expected association). They may therefore impact planning approval due to potential prot sharing benets or due
to stakeholder objections to externalities. The average site size (Area)
was 37 ha, with an average of 6 turbines (no. of turbines), on 56
meter high masts (hub height), with 56 meter blades (Rotor Diameter),
and producing 11 MW of electricity (Rated Outcap).
Creating indicators for the institutional process is more complex. We
disaggregate the process variables into three subcategories, one describing institutional aspects of the process, and two others characterizing
refusal reasons stated by local and national planning authorities. One refusal subcategory relates to ecological impacts, the second to human
welfare impacts. For the institutional subcategory, Pettersson et al.
(2010) suggests that the type of processes initiated (local, national),
the time delay between submission of applications and appeal and decision dates, and the deliberation periods of various planning authorities
(in days and capped at 2000 days), can provide a good proxy for the institutional process. Further, other works suggest that public participation in the planning process, inclusive public meetings, the presence of
third party groups, and the interaction of developers with the local community are key indicators of positive planning approval outcomes
(Devine-Wright, 2005b; Gross, 2007; Higgs et al., 2008; Loring, 2007;
Toke, 2005a, 2005b; Wolsink, 2006; Zoll, 2001). We therefore include
variables describing the types of approvals sought, variables describing
the durations of the various stages of the applications and appeals process, variables distinguishing involved stakeholders, project applicants,
and third party groups, as well as variables describing the type of appellant (individuals, project proposers, versus conservation or community
groups). Variables representing the recommendations of the 1st inspectors and the initial decisions of local authorities are also included as suggested by Haggett and Toke (2006) and Toke (2005a, 2005b) due to the
impacts that independent assessor recommendations or initial local decisions can have on WFs ultimate probability of approval. Pettersson
et al. (2010) also suggests the motivations of various involved agents
(in terms of refusal reasons) may also impact the planning outcome.
For instance, earlier work suggests refusals or objections related to ecological impacts (e.g. protected landscapes, nature areas, ecosystems, or
waterways) tend to dominate refusals related to human welfare impacts (e.g. negative externalities related to project scale, noise, or visual
impacts); (see Devine-Wright, 2005a; Dimitropoulos and Kontoleon,
2009; Jones and Eiser, 2009; Loring, 2007; Thayer and Freeman, 1987;
Wolsink, 2000, 2007b). The refusal reason variables we include are all
reasons stated by appellants during appeals for the projects in our
sample.
We observe that applicants on average take 68 days to submit an appeal to a local decision once it is made (Time2Ap1), and that this local
appeal process typically lasts 67 days up to a maximum of 734 days
(Appeal1Duration). The average total duration of the local approval
and appeal process (TotLocDuration) is 175 days. Further, the 1st inspector grants an average of less than 1% approvals (1st Inspect
Grant), 19% refusals (1st Inspect Refuse), with the remainder being approvals granted with conditions. Third, 22% of appeals are by applicants
(AppealApplicant), typically to conditions associated with approvals,
1% are by groups usually composed of local stakeholders who may or
may not be directly impacted (AppealGroup), 2% of appeals are offered
by directly impacted stakeholders (AppealAffectParty), and 18% are by
the 3rd parties (AppealThirdParty) such as environmental action
groups. Finally, 2% of proposals are removed from local authority consideration during the approval process (Bord Withdrawn), and local authorities refuse 28% of projects (LA refused).
The remaining process variables describe the reasons for refusals
stated by the local and national planning authorities. Note that the

15

sum of the refusal reason percentages does not equal to one because
not all projects are refused, projects may be refused for multiple reasons, or refusal reasons may not be reported, i.e. may not relate to
human or ecological impacts. In terms of ecological effects, the most
cited refusal reasons include impacts upon birds (REF Birds) or on
peat stability (REF Peat Stability) both at 2% of the 354 projects in our
sample. Refusal reasons related to environmental impact assessments
predicting ecological damage (REF EIA) or proximate to protected
Natura sites (REF Natura) each represent about 1% of the sample. The
other refusal reasons, including impacts on archeological sites which
are discussed later, each accounted for less than 1% of the sample. Refusal reasons related to human welfare impacts are much more common.
The most often cited refusal reason is related to visual externality (REF
Visual) at 17% of the sample. Next most common are reasons
related to conicts with existing strategic development plans (REF
StratDevPlan) at 12%, proximity to dwelling (REF Prox Dwelling) at
5%, negative implications for public safety (REF Public Safety) at 2%,
and potential aviation hazards or negative tourism impacts (REF Aviation, REF Tourism) each accounting for 1% of the sample.
3. Empirical Methodology
The goal of our empirical analysis is to determine how a large set of
predictor data X are associated with the probability of wind farm planning approval Y. The most direct analysis would be to simply apply
Probit to our predictors and obtain a set of parameters bRaw, e.g. Prob
Raw 
p 2
X0 b
Y 1jX
1= 2 et =2 dt . Unfortunately, there are several
problems with the data that preclude this direct analysis including variations in predictor distributional properties and magnitudes, the presence of outliers, and predictor collinearity.
To address these problems we sequentially employ six statistical
techniques, see the online technical Appendix B. First, in order to
allow more direct comparison of the relative effects of the predictors
X, we normalize all non-binary data using z-scores; this yields a predictor dataset x. The dependent variable Y is a binary term representing
planning approval or not, therefore this and other binary data are not
normalized. Second, in order to provide a context for our more complex
and restricted model described below, we rst estimate a simple ProbitAll model which includes all normalized predictor data ProbY 1jx
p 2
x0 bProbitAll 

1= 2 et =2 dt. Third, to control for the likely possibility of

collinearity among our normalized predictors and to reduce our large
set of predictors into a smaller orthogonal set we apply PCA. This yields
principal components (PC) ce
x; w which are conditional upon the normalized and binary predictor data e
x (the tilda indicates that a constant
term is included with the predictors dataset x), and a set of principal
component weights w; see Green (2000). Fourth, we employ Probit regression analyses relating the derived principal components ce
x; w to
the binary dependent variable Y; this yields a set of estimated parameters denoted b. Fifth, to recover the importance of the association among
predictors x (excluding the constant) and the dependent variable Y, we
employ a technique which combines the estimated parameters b and
component weights w and then sums the product of these parameters,
for each underlying predictor in x, and across the components observed
to be statistically signicant in step four. Finally, step six conducts robustness checks to verify our identied associations.
Importantly, there are three primary problems with using the principal components approach in step 3. First, the results are sensitive to
the scale of the data. The obvious solution is to standardize the independent variables as we have done. Second, the principal components are
not chosen based on any relations with the dependent variable Y one
is trying to explain. So a more important component (for a particular dependent variable) may be excluded in favor of a spurious high Eigen

16

T.M. van Rensburg et al. / Ecological Economics 111 (2015) 1222

value component. Our combined PCAProbit analysis in step 4 and associated step six robustness check results described in Section 4.3 addresses this issue. As a general rule one wants to use fewer
components than the number of variables in e
x. Using too many may
not eliminate the collinearity problem; alternatively, we cannot exclude
too many components otherwise we may miss important inuences.
Based on Scree plots we use L = 9 components out of the potential
K = 66. Finally, PCA makes it difcult to determine the inuences of
the individual variables underlying the components, since each component could include different transformations of all variables. Our importance approaches of Section 3.1 address this issue.
3.1. Calculating Predictor Importance
A unique aspect of our methodology is that we have multiple ways
to infer the importance of the underlying predictors in x, despite the
fact that they have been transformed into principal components c. Additionally, to deal with the limitation of PCA whereby important variables
may reside in low Eigen value components or may have opposite signed
inuences which cancel over components, we construct two importance
measures, net importance I and absolute importance ; see Jeserich et al.
(2012), Kelley (1998), Kelley et al. (under review), and Laepple and
Kelley (2013).
See the online technical Appendix B for specic details regarding the
calculation of these measures. However, briey consider that after identifying the principal components weight matrix w and obtaining the i =
L + 1 regression coefcients (including the constant as L + 1), the relationship among predictors x, PC weights w, Probit coefcients b, and the
predicted Y is,
^ b0
y

L
X



bi  wi;1  x1 wi;2  x2 wi;K  xK

i1

where i may range from 1 to L K components. Since each metric potentially appears in each component, i.e. across i in the above summation, to determine the importance of any metric in the overall
regression we need to sum the product of the regression coefcient
for the principal component and the principal component weight for a
predictor for all principal components included in the regression; crucially we only sum across principal component regression parameters
that were statistically signicant at the 99% level. Specically, the net
importance of any individual predictor xK is represented as IK.
IK

L
X



bi  wi;K  jt b;i j2:58 :

i1

The inequalities in Eqs. (3) and (4) are Boolean tests that take a value
of one or zero if the t-statistic for a particular component equals or exceeds the 1% signicance level. Importantly, since PC weights and regression coefcients can be positive and negative, using this method
identies only the net effect of potentially important drivers. The problem with this approach is that we may have important predictors that
have both positive and negative effects, and by summing across components we erroneously exclude them from the list of important predictors as they cancel out. To control for this we also consider the sum of
the product of absolute values of component weights and regression coefcients, i.e. our absolute importance measure.
eI
K

L
X



jbi j  jwi;K j  jt b;i j2:58 :

i1

Eq. (4) tells us about predictors that are important in an absolute

sense and are statistically signicant, but this measure will not tell us
the direction of the net inuence as in the former case. By considering
both of these importance measures we can more robustly identify

statistically important predictors and the direction of the effect of x on

our dependent variable Y.
3.2. Robustness Checks
In order to determine the robustness of our importance results we
perform two types of checks. Our baseline analysis employs the zscore normalized continuous variables, with principal component analysis, Probit regression analysis, and net importance scores. Our rst robustness check employs the absolute importance measure and retains
all other analysis methodologies. Any difference among the baseline
and the rst robustness check results indicates that predictor effects
may cancel out when summing across components. This could indicate
that our net measure in Eq. (3) may underestimate the importance of a
predictor.
A second robustness check is performed in order to address the second limitation of PCA, namely that important variables may reside in
low Eigen value components, i.e. in components beyond the L = 9
that are considered. To determine if this is the case, as many PCs are included in the Probit analyses that allow a full rank predictor matrix (in
this case L = 40 out of the possible 66). This allows importance scores to
capture any variable inuences in L N 9 components, and can allow us to
validate any anomalous importance score signs.
Our nal robustness check presumes that the linked PCAProbit
methodology we propose may only be an exploratory data analysis
technique similar to exploratory factor analysis (Ellram et al., 2013).
Under this interpretation, one can consider the importance score as a
measure to identify appropriate variables to include in a more basic
(Probit) regression using normalized predictors rather than principal
components. An added benet of this robustness check is that it gives
us a sense of the statistical relevance of the pseudo-T statistics calculated for importance scores. Ideally, variables meeting importance score
pseudo-T statistic signicance thresholds (N1.96) should match the variables meeting the Probit-Conrm analysis T statistic thresholds.
4. Results and Discussion
Our results are based on an analysis of 66 control, endowment, process, and project variables across 354 wind farm projects in the Republic
of Ireland between 1990 and 2011. This cross sectional database includes over 3300 observations that can be used to address our research
questions. The object of our analysis is to determine which individual
variables or variable categories were best able to account for WF project
planning approval. For brevity, the PCA analysis results (i.e. factor loadings), and PCAProbit regression analysis results (parameters for these
components) are omitted; however, the latter can be obtained from
the online technical Appendix B.
4.1. Simple Probit-all Model Results
Table 2 provides the results of a simple analysis associating planning
approval with all the normalized predictor data. The adjusted McFadden
R2 for this test is 0.66 and the variance ination test value is approximately 3, this suggests no collinearity. However, the mean absolute correlation among predictors is 0.1, the standard deviation is 0.14, and the
kurtosis exceeds 32; the latter suggests a rightward shifted (toward
1) correlation distribution. The Wilcoxon rank sum test, for the vector
of all absolute predictor correlations, indicates one can reject the hypothesis that the median correlation equals 0, or is greater than or
equal to 0.8; however one cannot reject the hypothesis that the median
correlation lies between 0.1 and 0.7. Overall this evidence suggests that
a high degree of predictor collinearity may be impacting these Probit-All
results. Nevertheless, despite the potentially inated standard errors
this implies, 10 variables appeared signicant at the 10% level or less;
these are reported above the thin line in the Table. Note that due to

T.M. van Rensburg et al. / Ecological Economics 111 (2015) 1222

Table 2
Simple Probit-All analysis associating all predictors to project approval success. Note,
county dummy variable results not reported. Mcfadden adjusted R2 equals 0.66.

17

they can be useful for providing a context for model comparisons with
the more complex and restricted PCAProbit model below.

Variable

Variable type

Regression
coeff.

T-stat

4.2. Calculated Importance Scores

AppealThirdParty (1 = yes)
REF archeology (1 = yes)
1st Inspect Refuse (1 = yes)
AppealApplicant (1 = yes)
REF StratDevPlan (1 = yes)
ApplicantSubsid. (1 = yes)
AppealAffectParty (1 = yes)
REF Prox Natura (1 = yes)
EIS (1 = yes)
REF Birds (1 = yes)
ApplicantCountry (1 = ROI)
AppealGroup (1 = yes)
REF Peat Stab (1 = yes)
REF Cum. Impacts (1 = yes)
Ap1Duration (days)
Rated OutCap (MW)
Bord Withdrawn (1 = yes)
ApplicantLtd (1 = yes)
1st Inspect Grant (1 = yes)
ApplicantIndividual (1 = yes)
WindEndowment (ave mps)
Closest town (km)
Windfarm Exten (1 = yes)
REF Tourism (1 = yes)
REF Flora (1 = yes)
REF Visual (1 = yes)
Area (ha)
Rotor Diameter (m)
LA refused (1 = yes)
REF Flooded area (1 = yes)
No. of turbines (#)
Contract Amend. (1 = yes)
TotLocDuration (days)
REF Prox Dwelling (1 = yes)
REF Habitat (1 = yes)
REF Aviation (1 = yes)
Permission (1 = yes)
REF Public Safety (1 = yes)
Time2Ap1 (days)
Perm Extension (1 = yes)
REF EIA (1 = yes)
Contract Exten. (1 = yes)
Hub height (m)

Process:Institut.
Process:Ecol. Effect
Process:Institut.
Process:Institut.
Process:Human Effect
Project:General
Process:Institut.
Process:Ecol. Effect
Project:General
Process:Ecol. Effect
Project:General
Process:Institut.
Process:Ecol. Effect
Process:Ecol. Effect
Process:Institut.
Project:Technology
Process:Institut.
Project:General
Process:Institut.
Project:General
Endowment
Control
Project:General
Process:Human Effect
Process:Ecol. Effect
Process:Human Effect
Project:Technology
Project:Technology
Process:Institut.
Process:Ecol. Effect
Project:Technology
Project:General
Process:Institut.
Process:Human Effect
Process:Ecol. Effect
Process:Human Effect
Project:General
Process:Human Effect
Process:Institut.
Project:General
Process:Ecol. Effect
Project:General
Project:Technology

3.64
13.25
4.16
5.72
2.82
3.29
2.20
2.74
1.32
1.99

5.26
4.82
3.78
3.60
2.15
2.02
1.99
1.69
1.66
1.64
1.57
1.39
1.35
1.35
1.25
1.25
0.98
0.97
0.87
0.80
0.77
0.67
0.62
0.54
0.49
0.49
0.38
0.37
0.34
0.27
0.26
0.24
0.19
0.19
0.18
0.16
0.12
0.11
0.10
0.10
0.10
0.06
0.05

To obtain information about the importance of underlying predictor

variables, we combine the regression parameters b with the factor loadings w using Eqs. (3) and (4). Table 3 reports these two measures of importance, the predictor variable categories, as well as a pseudo-T
statistic indicating the signicance of an importance score. This pseudo
T-statistic is calculated by dividing a variable's net importance score by
the standard deviation across all net scores. For brevity, Table 3 only reports importance scores for predictors which resided in signicant PCs.
Comparison of the absolute and net scores, and the pseudo-T statistics, can provide insights about the direction, magnitude, and signicance of a variable's impact. It should always be the case that the
absolute importance exceeds or equals in absolute magnitude the net
importance for a predictor. If the net score is smaller, this indicates
that a variable's inuence is having countervailing (positive and negative) impacts across multiple PCs that cancel when summing to recover
its score. This may indicate that the net score is under-estimating the
true importance of a variable. Alternatively, equality implies the variable has the same signed inuence across all PCs. We observe that the
net and absolute scores are very similar for most variables. Notable exceptions include LA refused, appeal by applicant, area, refusal reason visual, and rated output capacity. This may suggest that the net score may
under estimate the impact of these variables, and further analysis may
be warranted; see Section 4.3.
Based on the pseudo-T statistics, six variables can be said to be signicantly (10% or less) associated with project approval success. Of
these six, ve reect institutional processes and one represents endowments. The process variables relate to the duration of the rst appeal
process (with an importance score of 0.60), the decisions by local
parties (authorities (0.34) and inspectors (0.29)), and the identities
of the appellant (i.e. applicants (0.28) versus third parties (0.27)). The
positive appeal 1 duration effect suggests that longer local appeal processes are positively associated with more likely planning approval,
which may reect several features. Long appeal processes may allow
project applicants to provide or better craft secondary/revised applications, to respond to objections, or to interact with local stakeholders
more effectively. Further, a longer process may avoid knee jerk refusals
which may be more likely if planning authorities (who have a primary
responsibility to local constituents) are placed under time pressure or
have incomplete time to review relevant materials. The negative local
authority and 1st inspector refusal effects suggest that negative local decision can strongly follow a project through to the higher level appeal
process (national level) and still negatively inuence the nal probability of planning approval. The negative (positive) sign of the variable
reecting appeals by applicants (3rd parties) suggests applicants' appeals reduce the probability of planning approval, while appeals by
third parties increases the probability. However, the negative applicants' effect is later found to not be robust following the additional robustness checks described in Section 3.2. These two results together,
combined with the absolute importance score and robustness check results suggest that the appeal processes may be slightly more concerned
with impacts on project proposers compared to the objections by third
parties such as wind energy advocacy groups or environmental conservation groups. Importantly, the negative applicants' result would be
considered anomalous since typically applicants would be appealing refusals with the aim of increasing the probability of planning approval, so
a positive parameter is expected. However, third parties may object to
project approvals if they are conservation groups, or they may object
to refusals if they are wind energy proponents, thus negative or positive
effects are plausible. The positive importance score for appeals by third
parties suggests that inuences from wind energy support groups may
dominate objections of conservation groups. The wind endowment

2.09
3.50
1.80
6.00
0.26
0.35
3.67
1.71
9.64
1.42
0.20
0.15
2.89
4.35
2.89
0.47
0.10
0.16
0.38
3.10
0.09
0.83
0.05
0.26
2.40
1.19
0.50
0.21
0.04
0.51
0.40
0.24
0.02

Indicates 10% signicance or less.

the potentially upwardly biased standard errors, some signicantly correlated variables are likely to have been spuriously omitted.
First, no control or endowment variables appear signicant. However, four institutional process variables are signicant including the positive effects of appeals by third parties, appeals by applicants, and
appeals by affected parties; the latter two groups are not necessarily exclusive. This result likely captures project proposers and third party
wind energy advocacy groups attempting to positively inuence the
planning outcome after local refusals. A signicant negative effect is observed for the 1st inspector refuse. Next, three process variables
associated with ecological effects are signicant including the counterintuitive positive effects of REF archeology and REF Proximity to Natura,
and the negative effect of REF Birds. In addition to being counterintuitive, the REF archeology result is particularly unusual given the small
number of instances this refusal reason occurred; see Table 1. These
are likely spurious results caused by the misspecication associated
with this simple preliminary model. However, the signicant negative
ecological effect REF Birds is intuitively plausible. Also, a negative and
plausible process result associated with human effects occurs for REF
Strategic Development plan. Finally, two project variables appear significant including the positive effects of the general project variable applications by subsidiaries and inclusion of an EIS. Overall, the results for
this basic unrestricted test should be considered cautiously, however

18

T.M. van Rensburg et al. / Ecological Economics 111 (2015) 1222

Table 3
Calculated importance of predictors underlying statistically signicant (1%) PCs for Probit and z-score normalized data. Dependent variable is project approval success. McFadden adjusted
R2 for Probit with PCs and constant term equals 0.31.
Variable

Variable type

Net importance

Pseudo-T

Absolute importance

Appl1Duration (days)
WindEndowment (ave mps)
LA refused (1 = yes)
1st Inspect Refuse (1 = yes)
AppealApplicant (1 = yes)
AppealThirdParty (1 = yes)
Area (ha)
REF Visual (1 = yes)
REF StratDevPlan (1 = yes)
Rated OutCap (MW)
Hub height (m)
Closest town (km)
ApplicantIndividual (1 = yes)
Time2Ap1 (days)
ApplicantLtd (1 = yes)
Rotor Diameter (m)
REF Prox Dwelling (1 = yes)
No. of turbines (#)
EIS (1 = yes)
Permission (1 = yes)
TotLADuration (days)
REF Habitat (1 = yes)
REF EIA (1 = yes)
REF Cum. Impacts (1 = yes)
Bord Withdrawn (1 = yes)
AppealAffectParty (1 = yes)
REF Tourism (1 = yes)
Contract Amend (1 = yes)
Mean
SD

Process:Institut.
Endowment
Process:Institut.
Process:Institut.
Process:Institut.
Process:Institut.
Project:Technology
Process:Human Effect
Process:Human Effect
Project:Technology
Project:Technology
Control
Project:General
Process:Institut.
Project:General
Project:Technology
Process:Human Effect
Project:Technology
Project:General
Project:General
Process:Institut.
Process:Ecol. Effect
Process:Ecol. Effect
Process:Ecol. Effect
Process:Institut.
Process:Institut.
Process:Human Effect
Project:General

0.60
0.37
0.34
0.29
0.28a
0.27
0.20
0.20
0.16
0.12b
0.12
0.07
0.07
0.07
0.07
0.06
0.06
0.06
0.03
0.02
0.02
0.02
0.02
0.02
0.02
0.01
0.01
0.01
0.00
0.19

3.86
2.37
2.20
1.83
1.79
1.77
1.28
1.26
1.02
0.77
0.75
0.48
0.48
0.46
0.44
0.41
0.38
0.37
0.17
0.14
0.13
0.12
0.11
0.11
0.10
0.09
0.09
0.09

0.60
0.37
0.46
0.38
0.41
0.27
0.50
0.33
0.24
0.34
0.19
0.22
0.07
0.25
0.07
0.17
0.07
0.19
0.07
0.02
0.36
0.02
0.02
0.03
0.02
0.02
0.02
0.01
0.20
0.17

a
This negative importance score becomes signicantly positive (0.63) when additional PCs are allowed in the Probit analysis as a robustness check; note that this adjusted parameter is
used for the comparative statics described in the conclusion.
b
This insignicant negative value becomes marginally signicant and positive (0.01) when additional PCs are allowed in the Probit analysis during a robustness check; note that this
adjusted parameter is used for the comparative statics described in the conclusion.

variable also appears signicant. The positive endowment effect (0.37)

suggests that this variable may reect the positive inuence of high
protability on the probability of approval.
Although the pseudo-T statistics suggests that the project variables
may be insignicant, it is worth considering the importance scores for
variables identied to be signicant in earlier work. Of these, the most
important impacts follow from project area (0.20), rated output capacity (0.12), and hub height (0.12). The positive values for area and hub
height may suggest that these represent protability conditions which
positively correlate with planning approval success (Mudasser et al.,
2013). The negative output capacity variable may indicate that this variable reects the negative landscape externalities associated with large
project footprints. However, this effect is later found not to be signicant during robustness checks. Unreported results indicate that all
county dummy variable importance scores are insignicant, which suggests no systematic bias in outcomes based on the county of residence.
In order to determine if our more complex and restricted PCAProbit
model is superior to the unrestricted Probit-All alternative, we compare
adjusted R2s and apply, AIC, BIC, and likelihood ratio tests comparing
the models. Note that these two models are nested, albeit in a rather
complex manner; see Eq. (2) and online Appendix B. The adjusted McFadden R2 for the unrestricted Probit-All model is 0.66 and the value for
the PCAProbit model is 0.31, suggesting support for the unrestricted
model. The AIC for the Probit-All model is 224.7, while the value for
the PCAProbit model is 243.2, suggesting that the unrestricted model
may be slightly preferred. However, the BIC values are, respectively,
480.1 and 281.9 suggesting the PCAProbit model is strongly preferred.
Further, the log likelihood value for the unrestricted model is 46.4
while for the PCAProbit model it is 111.6; this yields a likelihood
ratio test value of 130.4 which with 56 degrees of freedom compares

to a 10% chi-squared critical value of approximately 43, thereby indicating one can fail to reject the null that the unrestricted model is superior.
Overall, this somewhat mixed evidence supports the unrestricted
Probit-All model. However, in conjunction with the substantial evidence of collinearity among the full set of non-normalized predictors
described in Section 4.1, we discount this evidence due to the high probability that the unrestricted model suffers from collinearity, spurious
correlation, and misspecication.
Nevertheless, despite the more precise estimates provided by the
more complex model, there are some similarities among the results of
the restricted and unrestricted models worth mentioning. The restricted
model results in Table 3 and the unrestricted model results in Table 2
suggest that there may be a positive effect of the institutional process
variables appeals by third parties, and a negative effect of 1st Inspector
refusals. Common human effect process variables include the negative
effects of refusals due to conicts with strategic development plans.
Note that there were no commonly identied ecological effect process
variables in the restricted model, or endowment or project variables
in the unrestricted model. These differences could indicate that the effects of refusal reasons related to ecological effects and environmental
impact assessments may be less important than suggested in the earlier
literature (i.e. they may be related to econometric problems which spuriously suggests their signicance as in the unrestricted model here).
And further the restricted model suggests that features such as wind endowments, and institutional process features associated with process
duration and local authority decisions, as well as certain project features
associated with applicants' identities and technology attributes, may be
more important than previously thought (i.e. they may be absent in the
unrestricted model or earlier literature due to collinearity and inated
standard errors).

T.M. van Rensburg et al. / Ecological Economics 111 (2015) 1222

Table 4
Probit-Conrm model robustness check results describing associations of project planning
approval and normalized predictors identied to have high importance. Mcfadden adjusted R2 equals 0.16.
Dependent variable: project approval success
Variable

Variable type

Regression
coeff.

T-stat

Ap1Duration (days)
1st Inspect Refuse (1 = yes)
LA refused (1 = yes)
REF StratDevPlan (1 = yes)
Area (ha)
AppealThirdParty (1 = yes)
Rated OutCap MW (MW)
AppealApplicant (1 = yes)
Rotor Diameter (m)
Time2Ap1 (days)
Hub height (m)
WindEndowment (ave mps)
REF Visual (1 = yes)

Process:Institut.
Process:Institut.
Process:Institut.
Process:Human Effect
Project:Technology.
Process:Institut.
Project:Technology
Process:Institut.
Project:Technology
Process:Institut.
Project:Technology
Endowment
Process:Human Effect

1.05
1.88
1.66
1.08
0.31
0.64
0.20
1.18
0.14
0.07
0.08a
0.04b
0.18

6.46
3.89
2.76
2.40
2.09
2.07
1.79
1.65
0.84
0.81
0.48
0.41
0.36

Note:
Indicates signicance at the 1%.
Indicates signicance at the (10%) level.
a
This marginally signicant negative value is marginally signicantly positive for PCA
Probit robustness checks allowing additional PCs in the analysis. Thus, this negative value
may be discounted during the comparative statics described in the conclusion.
b
This insignicant negative value is signicantly positive for both the PCAProbit Baseline
and robustness checks allowing additional PCs in the analysis. Thus, this negative value may
be discounted during the comparative statics described in the conclusion.

4.3. Probit-Conrm Robustness Checks for High Importance Variables

Table 4 reports the results of robustness check performing a Probit
analyses linking planning approvals directly (not through principal
components) to normalized predictor variables. Such a methodology
implicitly interprets the rst stages of our PCAProbit analysis as exploratory, i.e. identifying appropriate independent variables, while the nal
Probit-Conrm analysis of the predictors identied would be considered conrmatory. Predictors included are those identied to have
high pseudo-T statistics, large differences in net and absolute importance scores, or those which were identied in earlier studies but
which had pseudo-T values just on the other side of signicance. Including some presumably non-signicant predictors allows us to determine
if the pseudo-T threshold of 1.96 is an appropriate approximation of a
signicance threshold for importance scores.
The adjusted McFadden R 2 for this Probit-Conrm test is 0.16.
And, in all but a few cases the signs and signicances in Table 4
Probit parameters match those of Table 3 importance scores. The
exceptions occur for appeals by applicants and rated output capacity where in Table 3 we observe a negative importance score but in
Table 4 we identify a positive effect; and also for hub height and
wind endowment which displays the opposite pattern. These discrepancies are most likely related to the potential weakness of the
combined PCAProbit approach, namely that it can potentially distribute the inuence of a key variable over many (potentially low
Eigen value) components because predictors are grouped in components without regard to a particular dependent variable. As a result, our method of using as few principal components as possible
might cause us to miss the inuence of a variable in a high number
PC. To test the validity of this explanation, we re-perform the entire
analysis allowing 40 of the 66 potential PCs to be included in the
basic regression (the maximum allowing a full rank predictor matrix). Strictly speaking this is not a preferred way to perform an
analysis since including many PCs reduces statistical power, leaves
the potential for collinearity to remain among predictors, and increases the potential for spurious correlations; however this can
be used as a diagnostic test. When including 40 PCs, the importance
scores for appeals by applicants, output capacity, and hub height

19

become marginally signicantly positive. Other variable signs, including the positive wind endowment effect, are conrmed. Finally,
this approach suggests two additional variables may be signicant
including the positive effect of hub height, also identied in the
PCAProbit robustness check, and the negative effect of REF strategic development plan.
In order to determine if our more complex and restricted PCAProbit
model is superior to the less restricted Probit-Conrm alternative, we
again compare adjusted R2s and apply, AIC, BIC, and likelihood ratio
model comparison tests. The adjusted McFadden R2 for the less restricted Probit-Conrm model is 0.16 and the value for the PCA
Probit model is 0.31, suggesting support for the proposed restricted
model. Also the AIC for the Probit-Conrm model is 298.4, while
the value for the PCAProbit model is 243.2, and the BIC values are,
respectively, 348.7 and 281.9. Both suggest the PCAProbit model
is preferred. Finally, the log likelihood value for the unrestricted
model is 136.2 while for the PCAProbit model it is 111.6; this
yields a likelihood ratio test value of 49.3 which with 3 degrees
of freedom at 10% signicance compares to a critical value of approximately 0.115. This indicates one can reject the null that the unrestricted model is superior. Overall, this evidence consistently
supports the superiority of the PCAProbit model, and therefore
the relevance of the importance scores we derive from it.
5. Conclusions and Policy Implications
This paper investigates the inuence of control, endowment, process, and project attributes on the probability of wind farm planning
approval in the Republic of Ireland during 1990 to 2011. To date, little work has rigorously investigated the s inuences of these factors
on wind farm planning outcomes. A methodological contribution of
this paper is to provide a unique measure of predictor variable
importance.
Study ndings reveal that institutional process characteristics are
by far the most important determinants of planning success, especially the duration of the local authority deliberation processes, decisions of local authorities and inspectors, and the identities of the
appellants. Perhaps the most important project technology variables
are project area, rated output capacity, and hub heights. Wind endowments appear to have a marginal inuence. The most important/inuential process variable refusal reasons relate to strategic
development plans and to a lesser extent visual impacts, which is
consistent with previous literature (Devlin, 2005; Jobert et al.,
2007; Nohl, 2001; Wolsink, 2006, 2007a; Johansson and Laike,
2007; Toke et al., 2008; Warren and McFadyen, 2010). Interestingly,
control variables such as county of project residence and nearest
town do not appear to matter. And similarly, refusal reasons related
to proximity to dwellings or special areas of conservation (SACs) also
appear insignicant. This suggests that NIMBYism does not inuence
Irish wind farm planning outcomes. This is similar to the lack of evidence suggested by Devine-Wright (2005b), Warren et al. (2005),
and Johansson and Laike (2007) regarding project proximity on attitudes about wind farms. Importantly, one reason proximity to dwelling, town, or SAC may not matter is because minimum distance laws
as stated in wind farm planning guidelines (no formal laws exist at
this time) which may cause rms not to submit proposals for sites
that are too close to certain areas because of the low probability of
approval success. Thus, there may be a sample selection/self-selection
explanation that accounts for this result.
In terms of research question 2, different parties have different
objectives (in terms of planning approval), as well as different variables under their control. However, our ndings can inform government policy makers, project proposers, and local stakeholders and
residents. With respect to governments and the institutional process, if the general goal is to increase adoption of wind energy per
EU and US directives, policy makers should not impose excessively

20

T.M. van Rensburg et al. / Ecological Economics 111 (2015) 1222

binding time constraints on the duration of the application and appeal process. Although parties deserve timely consideration, our results suggest that severe time constraints could result in a bias
toward refusal perhaps due to the greater prevalence of negative
knee jerk decisions.
Project proposers would be well advised to seek out sites with high
wind endowments, despite proximity to dwellings and towns because,
although proximity is important to local stakeholders, it is associated
with relatively small to marginal effects on planning approvals compared to endowments. This is consistent with Butler and Neuhoff
(2008), Lee et al. (2009) and Mudasser et al. (2013). Of course excessive
proximity would be ruled out by minimum distance laws. Project technology features which project proposers might be advised to promote
including large project areas, high rated output capacity, and to a less
extent high hub height; this is consistent with Devine-Wright
(2005a), Dimitropoulos and Kontoleon (2009), Meyerhoff et al. (2010)
and Haaren, van and Fthenakis (2011). These positive predicted inuences suggest that planning authorities prefer larger, presumably
more protable, projects. Overall, these results suggest that WF policy
incentives might be better targeted toward large scale projects (or at
least large areas, numbers of turbines, and mast heights) which have
the greatest probability of approval success.
Two process results that are of major importance to project proposers are associated with the decisions of local authorities; this is consistent with Haggett and Toke (2006), Toke (2005a, 2005b). Namely, if
an initial application is refused by local authorities and/or 1st inspectors,
these effects may dominate all other positive effects produced by scaling
up the project and protability, or by appeals by proposers or supportive third party advocacy groups (for the latter two effects see below).
Awareness of this could motivate proposers to better craft their initial
applications or proposal revisions, or simply lead to more application
withdrawals following initial failures.
Our results also provide clear recommendations to the various
classes of concerned parties regarding the appeals process. Generally, appeals by applicants and third parties both positively inuence
the planning outcome, although appeals by applicants can be expected to have a larger impact than those by third parties. This means
that project proposers can expect that their objections (to refusals)
will receive at least as much attention as those of third parties; thereby allowing them to adjust their project risk assessments downward
in the presence of third party opposition groups. However, the overall positive third party effect suggests that wind energy promotion
groups are more inuential/present compared to environmentally
oriented opposition groups. Overall, these results can inform affected parties about the required level of effort to expend compared to
(from their perspective) opposition groups. For example, if there is
a third party resistance to a project, a project proposer would be
well advised to promote third party support groups coupled with
an appeal of the negative decision. However, in the absence of any
third parties, an applicant may have success by simply appealing a
negative local decision.
Finally, the results describing the importance of process variables
associated with refusal reasons are of particular relevance to locally
affected residents (see Kelley et al., 2012, for a similar contribution).
In general, refusal reasons have smaller importance/signicances
than the factors described above. However, refusal reasons based
upon conicts with the strategic development plan are signicant
and negative, while visual objections are marginally signicant and
negative; the latter is consistent with Landenburg and Dubgaard
(2007) and Meyerhoff et al. (2010). The greater effect of human welfare rather than ecological effect refusal reasons contrasts with the
evidence reported by Thayer and Freeman (1987), Devine-Wright
(2005a), Jones and Eiser (2009) and Wolsink (2007b). The results
here suggest that plausible objections based on strategic development plans and visual externalities may be the best (the only) effective resistance vehicles. Other human effect refusal reasons such as

proximity to dwellings, or ecological effects refusal reasons related

to protected habitats or violations of environmental standards are
not likely to be effective. Perhaps since projects which may be plausibly obstructed for these reasons may not have been submitted in
the rst place.
In terms of practical implications, we can convert Table 3 PCAProbit
net importance scores, and Table 4 Probit-Conrm estimated coefcients for the variables previously discussed, into expected changes in
the probability of approval. This is a form of numerical comparative statics. Assume a baseline wind project in Donegal country, for which an
ROI resident Ltd. corporation seeks a new project permission, where
all institutional process variables, except appeal 1 duration, are set to
zero, and with the former variable and all other continuous project technology variables being set to the minimum value observed across projects, or the mean value if the minimum equals zero; see Table 1. One
can then recover the baseline predicted approval probabilities for the
PCAProbit (Probit-Conrm) models with (Y = 1|x) = (xINet)
((Y = 1|xConrm) = (xConrmbConrm)) where represents the cumulative normal probability distribution; see Green (2000) pages 773 and
1062. Note, for the discussion which follows a term in parentheses
will represent the Probit-Conrm prediction if available, i.e. there was
a signicant parameter in Table 4, or an implausible result could not
be discounted, see footnotes a and b. Also note that the two anomalous
Table 3 importance scores are replaced with the robustness check importance scores, see footnotes a and b. The baseline project's probabilities of approval are p = 0.26 (p = 0.33). We next report the results of
experiments modifying key signicant predictor variables, and a percent change represents a change relative to these values.
First, in terms of process variables associated with human welfare effects, introducing refusal reasons related to strategic development plans
or visual objections are associated with reductions in the probability of
project approval by 15% and 18% (7% and 12%), respectively. In terms of
institutional process variables, extending the 1st appeal duration by one
standard deviation, approximately 140 days, is associated with an increase in the probability of approval by 51% (10%). Further, if the local
authority or 1st inspector refuses to provide an approval this reduces
the ultimate probability of approval by 31% or 26% (91% or 95%), respectively. Finally, if an applicant or third party appeals a local authority refusal, the probability of ultimate approval increases by, respectively, 48%
or 25% (3% or 21%).
Second, locating a project on a site with a one standard deviation
higher wind endowment, approximately 1 mps out of a total ROI
range of 5 mps, is associated with an increase in the probability of approval by 21% (2%). Third, project technology attributes associated
with increases in approval probability include increases in project
area, a one standard deviation increase of about 80 ha results in an
18% (15%) approval probability increase. A one standard deviation increase in hub height, approximately 20 m, predicts a 12% increase in approval probability. Finally, a one standard deviation increase in output
capacity, approximately 12 MW, increases the approval probability by
1% (11%).
It should be noted that the parameters used to derive these predictions are obtained within the context of ROI, therefore they may
have limited international generalizability and should be applied
outside ROI with caution. Additionally, given the model comparison
results summarized at the end of Section 4.3 suggesting the superiority of the restricted PCAProbit model over the Probit-Conrm
model, some of the more extreme predictions of this latter model
(in parentheses) should be discounted toward the values suggested
by the PCAProbit approach. Any differences in the predictions
among these two models are likely related to any remaining collinearity or misspecication problems the Probit-Conrm approach
cannot resolve. Finally, there is the possibility that non-linear effects
may be present for the continuous variables, so particularly large
predicted changes in approval probability should be considered cautiously; this is a topic of future work.

T.M. van Rensburg et al. / Ecological Economics 111 (2015) 1222

Appendix A

Appendix B. Supplementary Data

Supplementary data to this article can be found online at http://dx.
doi.org/10.1016/j.ecolecon.2014.12.012.

Table A.1
Detailed variable descriptions.
Variable

Description

Project Approval (1 =

Whether proposed wind farm project approved via

any path

yes)
Closest town (km)
WindEndowment (ave
mps)
Appeal1Duration (days)

1st Inspect Grant (1 =

Distance of closest town/village to proposed site

Ave. wind speed at heights (50 m, 75 m, 100 m) at
proposed site
Duration of local appeal process from submission to
disposal
Time from recipt of local decision until an appeal is
submitted
Duration of local process from submission to appeal to
national
Application approved by 1st Inspector

yes)
1st Inspect Refuse (1 =

Application refused by 1st Inspector

yes)
AppealApplicant (1 =

Appeal by Applicant

yes)
AppealGroup (1 = yes)
AppealAffectParty (1 =

Appeal by affected external group, e.g. community

Appeal by party directly affected

Time2Ap1 (days)
TotLocDuration (days)

yes)
AppealThirdParty (1 =
yes)
Bord Withdrawn (1 =
yes)
LA refused (1 = yes)
REF archeology (1 = yes)
REF Birds (1 = yes)
REF Cum. Impacts (1 =
yes)
REF EIA (1 = yes)
REF Flooded area (1 =
yes)
REF Flora (1 = yes)
REF Habitat (1 = yes)
REF Peat Stability (1 =
yes)
REF Prox. Natura (1 =
yes)
REF Aviation (1 = yes)
REF StratDevPlan (1 =
yes)
REF Prox. Dwelling (1 =

Appeal by third party such as environmental

non-prot
AnBordPleanala application or appeal withdrawn
Local Authority refused application
Refusal due to archeological ndings
Refusal due to protection of bird species
Refusal due to cumulative impacts on the locality
Refusal due to unfavorable environmental impact
assessment
Refusal due to ooding risk
Refusal due to protection of local ora
Refusal due to habitat protection
Refusal due to peat instability, i.e. is present or will be
caused
Refusal due to proximity to Natura site
Refusal due to interference with aviation
Refusal due to conict with local strategic
development plan
Refusal due to proximity to dwellings

yes)
REF Public Safety (1 =

Refusal due to public safety concerns

yes)
REF Tourism (1 = yes)
REF Visual (1 = yes)
ApplicantIndividual (1

Refusal due to interference with local tourism activity

Refusal due to visual impact
Applicant is individual

= yes)
ApplicantLtd (1 = yes)
ApplicantSubsid. (1 =

Applicant is a limited company

Applicant is subsidiary company

yes)
ApplicantCountry (1 =

Applicant country

ROI)
Contract Amend. (1 =
yes)
Contract Exten. (1 = yes)
EIS (1 = yes)

Amendment of developer contract (# turbines, area,

etc) sought

Extension of contract duration sought

Environmental impact statement completed with
application
Perm Extension (1 = yes) Extension of permission (to build) duration sought
Permission (1 = yes)
1st time permission sought
Windfarm Exten (1 =
Extension of existing windfarm (area, # turbines etc.)
sought
yes)
Area (ha)
Hub height (m)
No of turbines (#)
Rated OutCap (MW)
Rotor Diameter (m)

21

Area in hectares of proposed site

Average hub height of proposed turbines
Number of proposed turbines
Rated output capacity in megawatts of proposed site
Rotor diameter of proposed turbines

References
Alvarez-Farizo, B., Hanley, N., 2002. Using conjoint analysis to quantify public preferences
over the environmental impacts of wind farms. An example from Spain. Energy Policy
30 (2), 107116.
Bergmann, A., Hanley, N., Wright, R., 2006. Valuing the attributes of renewable energy investments. Energy Policy 34 (9), 10041014.
Butler, L., Neuhoff, K., 2008. Comparison of feed-in tariff, quota and auction mechanisms
to support wind power development. Renew. Energy 33 (8), 18541867.
Delmas, M.A., Montes-Sancho, M.J., 2011. US state policies for renewable energy: context
and effectiveness. Energy Policy 39 (5), 22732288.
Devine-Wright, P., 2005a. Beyond NIMBYism: towards an integrated framework for understanding public perceptions of wind energy. Wind Energy 8, 125139.
Devine-Wright, P., 2005b. Local aspects of UK renewable energy development: exploring
public beliefs and policy implications. Local Environ. 10 (1), 5769.
Devlin, E., 2005. Factors affecting public acceptance of wind turbines in Sweden. Wind
Eng. 29, 503511.
Dimitropoulos, A., Kontoleon, A., 2009. Assessing the determinants of local acceptability
of wind-farm investments: a choice experiment in the Greek Aegean Islands. Energy
Policy 37, 18421854.
Ellis, G., SQW, 2012. A Review of the Context for Enhancing Community Acceptance of
Wind Energy in Ireland. SQW Pty Ltd and Queen's University Belfast, Belfast.
Ellram, L.M., Tate, W.L., Petersen, K.J., 2013. Offshoring and reshoring: an update on the
manufacturing location decision. J. Supply Chain Manag. 49, 1422.
Eltham, D.C., Harrison, G.P., Allen, S.J., 2008. Change in public attitudes towards a Cornish
wind farm: implications for planning. Energy Policy 36 (1), 2333.
Green, W., 2000. Econometric Analysis. McMillan, New York, NY.
Gross, C., 2007. Community perspectives of wind energy in Australia: the application of a
justice and community fairness framework to increase social acceptance. Energy Policy 35, 27272736.
Haaren, van, R., Fthenakis, V., 2011. GIS-based wind farm site selection using spatial
multi-criteria analysis (SMCA): evaluating the case for New York State. Renew. Sustain. Energy Rev. 15 (7), 33323340.
Haggett, C., Toke, D., 2006. Crossing the great divide using multi-method analysis to understand opposition to windfarms. Public Adm. 84, 103120.
Higgs, G., Berry, R., Kidner, D., Langford, M., 2008. Using IT approaches to promoted public
participation in renewable energy planning: prospects and challenges. Land Use Policy 25, 596607.
Horst, van der, D., Toke, D., 2010. Exploring the landscape of wind farm developments;
local area characteristics and planning process outcomes in rural England. Land Use
Policy 27, 214221.
Jeserich, N., Kelley, H., Toft, G., Cole, G., 2012. Drivers of high-growth businesses across US
states: policy implications. Kauffman Foundation Working Paper.
Jobert, A., Laborgne, P., Mimler, S., 2007. Local acceptance of wind energy: factors of success identied in French and German case studies. Energy Policy 35, 27512760.
Johansson, M., Laike, T., 2007. Intention to respond to local wind turbines: the role of attitudes and visual perception. Wind Energy 10, 435451.
Jones, C.R., Eiser, J.R., 2009. Identifying predictors of attitudes towards local onshore wind
development with reference to an English case study. Energy Policy 37 (11),
46044616.
Kelley, H., 1998. The inuence of expectations biases on asset prices: a test with marketexperiment data. Indiana University Working Paper.
Kelley, H., van Rensburg, T., Yadav, L., 2012. A micro-simulation evaluation of the effectiveness of an Irish grass root agri-environmental scheme. Land Use Policy 31,
182195.
Kelley, H., van Rensburg, T., Jeserich, N., 2014. Determinants of demand for recreational
walking trails in Ireland (under review).
Laepple, D., Kelley, H., 2013. Understanding the uptake of organic farming: accounting for
heterogeneities among Irish farmers. Ecol. Econ. 88, 1119.
Landenburg, J., Dubgaard, A., 2007. Willingness to payoff reduced visual disamenities
from off-shore wind farms in Denmark. Energy Policy 35 (8), 40594071.
Le Gourieres, D., 2014. Wind Power Plants: Theory and Design. Pergamon Press, New
York.
Lee, A., Chen, H., Kang, H.-J., 2009. Multi-criteria decision making on strategic selection of
wind farms. Renew. Energy 34 (1), 120126.
Longo, A., Markandya, A., Petrucci, M., 2008. The internalization of externalities in the production of electricity: willingness to pay for the attributes of a policy for renewable
energy. Ecol. Econ. 67, 140152.
Loring, J.M., 2007. Wind energy planning in England, Wales and Denmark: factors
inuencing project success. Energy Policy 35 (4), 26482660.
Maillebouis, C., 2003. Nimby ou la colre des lieux. Le cas des parcs oliens (Nimby or the
angst of the gods. The case of French projects for windmills). Nat. Sci. Soc. 11,
190194.
Meyerhoff, J., Ohlib, C., Hartjea, V., 2010. Landscape externalities from onshore wind
power. Energy Policy 38 (1), 8292.
Mguez, J.L., Lpez-Gonzlez, L.M., Sala, J.M., Porteiro, J., Granada, E., Morn, J.C., Jurez,
M.C., 2006. Review of compliance with EU-2010 targets on renewable energy in
Galicia (Spain). Renew. Sust. Energ. Rev. 10 (3), 225247.

22

T.M. van Rensburg et al. / Ecological Economics 111 (2015) 1222

Mudasser, M., Yiridoe, E.K., Corscadden, K., 2013. Economic feasibility of large community
feed-in tariff-eligible wind energy production in Nova Scotia. Energy Policy 62,
966977.
Navrud, S., Braten, K., 2007. Consumers' preferences for green and brown electricity: a
choice modeling approach. Rev. Econ. Polit. 117 (5), 795811.
Nohl, W., 2001. Sustainable landscape use and aesthetic perception preliminary reections on future landscape aesthetics. Landsc. Urban Plan. 54 (14), 223237.
Pettersson, M., Ek, K., Sderholm, K., Sderholm, P., 2010. Wind power planning and permitting: comparative perspectives from the Nordic countries. Renew. Sustain. Energy
Rev. 14 (9), 31163123.
Sahu, B., Hiloidhari, M., Baruah, D., 2013. Global trend in wind power with special focus on
the top ve wind power producing countries. Renew. Sust. Energ. Rev. 19, 348359.
Thayer, R.L., Freeman, C.M., 1987. Altamont: public perceptions of a wind energy landscape. Landsc. Urban Plan. 14, 379389.
Toke, D., 2005a. Community wind power in Europe and in the UK. Wind Eng. 29, 301308.
Toke, D., 2005b. Explaining wind power planning outcomes: some ndings from a study
in England and Wales. Energy Policy 33 (12), 15271539.
Toke, D., Breukers, S., Wolsink, M., 2008. Wind power deployment outcomes: how can we
account for the differences? Renew. Sustain. Energy Rev. 12 (4), 11291147.
Valentine, S.V., 2011. Sheltering wind power projects from tempestuous community concerns. Energy Sust. Dev. 15 (1), 109114.

Warren, C.R., McFadyen, M., 2010. Does community ownership affect public attitudes to
wind energy? A case study from south-west Scotland. Land Use Policy 27, 204213.
Warren, C.R., Lumsden, C., O'Dowd, S., Birnie, R.V., 2005. Green on green: public perceptions of wind power in Scotland and Ireland. J. Environ. Plan. Manag. 48, 853875.
Wolsink, M., 2000. Wind power and the NIMBY-myth: institutional capacity and the limited signicance of public support. Renew. Energy 21, 4964.
Wolsink, M., 2006. Invalid theory impedes our understanding: a critique on the persistence of the language of NIMBY. Trans. Inst. Br. Geogr. 31 (1), 8591.
Wolsink, M., 2007a. Planning of renewables schemes: deliberative and fair decisionmaking on landscape issues instead of reproachful accusations of non-cooperation.
Energy Policy 35 (5), 26922704.
Wolsink, M., 2007b. Wind power implementation: the nature of public attitudes: equity
and fairness instead of backyard motives. Renew. Sust. Energ. Rev. 11 (6),
11881207.
Yadav, L., van Rensburg, T., Kelley, H., 2012. A comparison between the conventional stated preference technique and an inferred valuation approach. J. Agric. Econ. 64 (2),
405422.
Zoll, R., 2001. Energiekonikte. Problemuebersicht und empirische Analysen zur
Akzeptanz von Windraftanlagen. Pol. Verhalt. 4.