Understanding the Amenity Impacts of Wind Development on an International Border

Martin D. Heintzelman, Richard J. Vyn, and Sarah Guth∗ January 13, 2016

DRAFT WORKING PAPER: DO NOT CITE OR QUOTE

Abstract Wind energy developments are often controversial. Concerns are often raised about negative impacts on local communities, including impacts on property values. Some of these negative impacts may be oset by compensatory payments made by wind developers. Communities often also have a say in approving the development or setting parameters which might dictate terms such as the density of towers or minimum setbacks from property borders or homes. However, if the development is near a border between municipalities, states, or even countries, it is often the case that one or more jursidictions will not have an opportunity to set such rules or demand compensation, but will, nonetheless, face some costs or impacts from the development. We explore exactly this situation at the border between Canada and the United States in the Thousand Islands region where a wind farm is currently operating on the Canadian border island of Wolfe Island. Using a parcel-level hedonic analysis of property sales transactions, we nd that properties in NY with a view of and/or in close proximity to the turbines signicantly depreciated in value after construction while parcels in Canada saw no signicant change in their value.

∗Heintzelman (corresponding author) is Associate Professor and Fredric C. Menz Scholar of Environmental Economics, School of Business, Clarkson University, P.O. Box 5790, Pots- dam, NY 13699, [email protected]. Vyn is Associate Professor, University of Guelph, Ridgetown Campus. Sarah Guth is Research Assistant, Harvard University. This research was supported by the Fredric C. Menz endowment fund for Environmental Economics at Clarkson University. Background research was conducted while Guth was participating in Clarkson University's Research Experience for Undergraduates (REU) program, supported by NSF Grant No. EEC-1359256. Additional research assistance was provided by Brittany Berry at the University of Guelph and Chuan Tang at Clarkson University. Some GIS analysis was provided by Adam Bonnycastle, also of the University of Guelph. We are indebted to seminar respondents at the University of New Hampshire and the 2015 Biennial Meeting of the Association for Canadian Studies in the United States for helpful comments and suggestions.

1 1 Introduction

Renewable energy sources are a steadily increasing portion of our global energy mix. Such energy sources are a global public good - by substituting for more pollution-intensive fossil-fuel sources they reduce global pollution of both crite- ria pollutants, such as NOx, SOx, Mercury, and others, as well as greenhouse 1 pollutants like CO2. The benets of these reductions are generally spread over a large area and, in the case of greenhouse gases, over the entire planet. The costs of these reductions, however, are more likely to fall on a much smaller ge- ographic area. In some cases, in fact, renewable energy facilities can be thought of simultaneously as global public goods and local public bads. As evidence of this, siting new renewable energy facilities, particularly wind farms, is often controversial, with local governments and/or activists putting up sti resistance. This can be true even in areas with populations which are generally supportive of environmental issues, as in the case of the Cape Wind development o the coast of Massachusetts2. Common local concerns about wind developments include visual and aural disamenities, potential human health impacts, and impacts on wildlife. The visual disamenities cited by wind opponents stem from the introduction of large man-made structures into the landscape. At the most basic level, such a change to the landscape can be o-putting to residents of an area, who may have bought their property under the assumption that the landscape would remain unchanged. More specically, opponents often cite the impacts of the large (depending on the facility) array of blinking red lights which sit atop the wind turbine hubs, as well as shadown icker which occurs when the turning blades are between the sun and a home, resulting in rhythmically moving shadows. The rst aect may be amplied when the view of the turbines is over water, as this clears the visual eld of other features which would otherwise obstruct or reduce such views. Noise impacts are also often cited by homeowners opposed to wind development. Of particular concern is low frequency noise which dissipates more slowly over distance (Bolin et al., 2011). Many people have complained about serious health impacts in relation to this low frequency noise. There have been some studies which nd evidence of health impacts, in particular sleep deprivation and psychological stress, but larger summary studies have found limited evidence linking the noise itself to these impacts (Council of Canadian Academies, 2015). Instead, as found by a recent Health Canada study, wind turbine noise is more likely to cause increased levels of annoyance rather than health impacts3. Regardless of scientic evidence, the perception of health impacts remains, and these perceived amenity and health impacts are likely to be reected in property values as bids for properties in close proximity to

1The exact emissions reductions from a wind facility depends very much on what other energy sources are displaced, the focus of Kane et al. [2013]. 2The Cape Wind project is a proposed oshore wind facility in Nantucket Sound. It has met strict opposition from many local landowners and, most famously, the Kennedy family. 3Health Canada. 2014. Wind turbine noise and health study: Summary of results. Avail- able at: http://www.hc-sc.gc.ca/ewh-semt/noise-bruit/turbine-eoliennes/summary-resume- eng.php, accessed December 11, 2015.

2 wind turbines will be reduced. Acting counter to these negative impacts are some benets which often ac- crue to local landowners and communities. First, individual landowners who are able to let their land out to wind energy companies stand to gain from streams of rental payments from developers. These contracts vary across both sites and landowners in magnitude and the allocation of risk (some contracts include xed payments without regard to electricity production whereas others include payments which are a function of electricity output, transferring some risk to landowners). In addition, developers often compensate communities through payments-in-lieu-of-taxes (PILOTs). These payments help to mitigate negative impacts in the local community and, again, vary in size and structure across sites. Payments to communities will also be reected in property values as these payments could result in either lower local taxes or increased local services, or a combination of both (Kahn, 2013). Payments to landowners are unlikely to broadly impact property values, but, if transferable, would certainly increase the value of parcels that include wind leases. Weber et al. [2013] note that energy royalties, including wind turbine lease payments, are resulting in substantial payments to farms across the United States It is also possible that the development of wind energy will result in increased economic activity through direct employment impacts and indirectly as payments made to individual landowners and the municipality ow through the local economy. Brown et al. [2012] nd signicant and positive economic impacts of wind facilities on personal income and employment, while Munday et al. [2011] nd less denitive results on economic impacts. Any positive economic impacts would likely also be reected in property values. Another issue in public acceptance of wind development is public involvement in the development/approval process. If a community feels that they have not had sucient input to the approval process, they are more likely to oppose and ght the development. In addition, if the project moves forward despite substantial opposition, this may lead to bad feelings and drive landowners out of the area, possibly lowering property values in the process. On the other hand, if the community is able to be actively involved in the process, they may have a stronger hand in negotiating better compensation for their community in addition to aecting details about the design of the project to their advantage.4 As an extreme form of this, if the development happens near a border between communities, but is wholly contained within one community, the community without the development is likely to be held out of the approval process, and will also not receive compensation from the developer. These neighboring communities, in other words, will bear some of the cost of the project with little prospect to receive any benets. All of these issues may be further complicated by the presence of second- home or vacation home owners. These owners are likely to have more elastic preferences regarding changes in the amenities surrounding their property. Since

4See Baxter, Morzaria, and Hirsch [2013] , Groth and Vogt [2014] , Ladenburg et al. [2013], Lindén et al. [2015], and Petrova [2013]for thorough discussions of the determinants of and strategies for social acceptance of wind energy.

3 they are not tied as closely to a particular region because of concerns about employment, they are more likely to sell their property, or less likely to buy a new property due to an adverse change in amenities. In addition, to the extent that additional local services, such as improved school systems, are provided through the use of PILOTs, they are less likely to be able to take advantage of these improvements. Finally, they may also not have as much of a voice in the approval process, and may feel like an outsider as a result. All of these factors make it possible that properties that are primarily used as second or vacation homes, or communities with a large share of parcels in this use, will face even larger changes in property values as a result of wind farm developments. This paper explores the issues discussed above using a unique scenario in which a large wind farm was constructed on the Canadian island of Wolfe Island in the St. Lawrence River at the border between Canada and the United States. Wolfe Island, which is the largest island in the Thousand Islands region, is situated at the entrance of the St. Lawrence River in Lake Ontario, directly across the river from the community of Cape Vincent in the state of New York. The Wolfe Island wind farm was developed by Canadian Hydro Developers, which initially submitted a proposal for construction of this wind farm on the western half of the island in July 2005 (Keating, 2006). The ocial plan and zoning bylaw amendments necessary to allow this project to move forward were passed by council in November 2006, and the project was ocially announced on the Wolfe Island website in April 2007. Construction of the 86-turbine, 197.8 MW facility began in May 2008 and was completed in June 2009, at which time the wind farm became operational (Ontario Power Authority). We examine for impacts of this wind farm on property values on both sides of the border using data from both Canada and the United States on property sales in close proximity to the wind turbines. On the American side, we focus on Jeerson County, which sits at the northern edge of New York and borders both the St. Lawrence River and Lake On- tario. Throughout the past decade, wind energy has divided public opinion in the county. The region has been the targeted site for several recent American wind facility proposals, including in the Town of Cape Vincent and in the Town of Hounseld, on Galloo Island in Lake Ontario, all of which have been highly controversial. Newspaper coverage and letters to the editor in the New York media regarding the Wolfe Island facility clearly expressed Jeerson County residents' opposition to turbines due to negative aesthetic impacts. A Cape Vincent jour- nalist feared that the turbines will take away [the] image and. . . beauty of [his] township, deterring prospective seasonal residents who contribute so much in taxes and expertise (Radley, 2009). A Chaumont resident characterized the wind farm as blight on landscape (Lynne, 2009). Finally, a seasonal resident of Chippewa Bay described the waterfront view of facility nighttime lighting as a jolt to the entire landscape and to [his] mind, . . . like a jab in the ribs (Quarrier, 2009). Similar sentiments have been expressed by residents of Wolfe Island, where the construction of this wind farm has generated considerable controversy and

4 public opposition. Opponents of the wind turbines have expressed concerns regarding the industrialization of this rural community and how the turbines forever change the landscape into something that doesn't t here (Fast et al., 2015). As with Cape Vincent, there are a considerable number of seasonal residences on Wolfe Island, many of which are waterfront properties. According to Fast et al (2015), summer cottages comprise about one-third of all residences on the island. As such, visual amenities play a signicant role in the value of these properties, and owners of these properties have expressed concerns regarding potential negative impacts on property values arising due to the visual disamenities associated with wind turbines. As one seasonal resident stated, why would I want to live there [with the turbines]? (Fast et al 2015). In one case, property owners brought an appeal to Ontario's Assessment Review Board to have the assessed value of their waterfront property reduced due to the devaluation caused by the wind turbines. This appeal was ultimately rejected due to a lack of evidence of negative impacts. But this case highlights the underlying concerns that exist among residents of Wolfe Island regarding impacts of wind turbines. However, not all residents of Wolfe Island were opposed to this project. As evident from interviews conducted by Fast et al (2015), there are a considerable number of area residents that were supportive of the project and of wind energy in general. It is interesting to note that similar concerns and issues were raised by residents on both sides of the border despite the fact that public meetings and open houses were held for residents on the Canadian side throughout the application and development process. Public open houses for this project were rst held in March of 2006, only a few months after the project was initially proposed. In October 2006, a public meeting was held to consider a proposed zoning bylaw amendment applicable to all Wolfe Island lands optioned for a wind plant zone (this amendment was passed by council the following month). In March 2007, public open houses were held that included maps indicating 86 turbine locations. Overall, public consultation focused only on residents of Wolfe Island, while residents of Cape Vincent had little to no involvement in this process. This dierence in the level of involvement could potentially contribute to a dierence in the nature of the resulting impacts on either side of the border. This study area is ideal for examining many of the issues discussed above. We have homes (in Canada) in very close proximity to the facility, but where residents had some knowledge of and, perhaps, some input into the project before it was built. The Township of Frontenac Islands, which includes Wolfe Island, also receives C$645,000 per year in payments from the developer, TransAlta5. In addition, we also have American homes with substantial views of the turbines, but from a distance, whose owners had no input to the project and whose communities receive no compensation from the developer. Many of the Ameri- can homes with views of the turbines, in particular, are also vacation or second homes rather than primary residences. We use a hedonic analysis to examine and compare how property values on

5TransAlta acquired Canadian Hydro Developers in 2009.

5 both sides of the border have been impacted by the Wolfe Island wind turbines. In this analysis, a dierence-in-dierences approach is used to compare transaction prices before and after approval or construction of the project as well as between homes which can and cannot view the turbines, or are at varying distances to the turbines. We employ xed eects to mitigate potential omitted variables bias as well as controls for property market trends and seasonality of prices. We nd robust evidence of negative property value impacts on the NY side after construction of the turbines for homes in close proximity to the turbines and/or with a view of the turbines. In contrast, we do not nd evidence of signicant negative impacts on the Canadian side.

2 Literature Review

There is a substantial and growing literature that examines property value impacts of wind development, focusing primarily on Canada, the United States, and Europe. This literature reveals mixed conclusions, with some studies nding no signicant impacts on property values, while others nd signicant negative impacts. Ben Hoen and his co-authors, in a series of studies looking at a large number of wind facilities across the United States, has consistently found no impact of wind turbines on property values (Hoen et al., 2011, 2015). These studies are impressive because of the large sample sizes which come from study- ing multiple sites around the country. However, by estimating an average impact across all included sites, this may obscure signicant impacts at individual sites. In fact,Heintzelman and Tuttle [2012] nd signicant negative impacts at two sites in northern New York but no impacts at a third site, which suggests that the impacts of turbines on property values can vary signicantly across dierent sites. Hence, combining multiple sites in one study can avoid the issue of low numbers of observations that has plagued prior wind turbine studies but does not permit observing dierences in impacts from one site or region to another. There are a number of other studies that also nd no signicant impacts of wind turbines on property values. Most recently, Lang et al. [2014] use a large dataset of transactions in Rhode Island to nd no signicant property value impacts of small wind facilities with either only a single turbine or a few small turbines. Laposa and Mueller [2010] nd no eects from the announcement of a proposed wind farm in Colorado, but that facility was never built so it dicult to interpret these results in the context of completed wind facilities. Shultz et al. [2015] examine the impacts of wind turbines on agricultural land values, and nd no impact. A limitation of this study, however, is that it uses assessed values rather than transaction values. As assessed values reect only the expert opinion of the local assessor, it is hard to know how accurate this analysis would be in assessing relatively new wind facilities. Sims and Dent [2007] and Sims et al. [2008] nd very limited evidence of turbine impacts, but use very small samples from sites with other confounding factors. Finally, Vyn and Mccullough [2014] study a large wind farm in rural Ontario and nd no evidence of signicant impacts on nearby property values. They do however point to the possibility

6 that impacts may occur in other contexts. This may particularly be the case in the province of Ontario where wind energy development has become increasingly controversial in recent years, and the concerns of impacts on health and property values have become prominent issues in the public forum. Heintzelman and Tuttle [2012] is the rst study that found signicant negative impacts from wind turbines. Importantly, however, they nd impacts in only 2 of 3 counties in their study area, suggesting that local context matters a great deal in determining the impact. Subsequently, a number of other studies have found signicant negative impacts. Sunak and Madlener [2012] implement a combination of spatial xed eects methods and Geographically Weighted Re- gression (GWR) and nd consistent evidence that proximity to wind turbines negatively impacts property values in a region of Germany. The GWR nds substantial variation in these impacts over space. Likewise, Jensen et al. [2014] and Gibbons [2015] both nd signicant negative impacts of wind turbines and, in the case of Jensen et al. [2014], both visual and aural disamenities are separately found to have negative impacts. Taken together, this growing literature suggests that there may not be any single, global answer to the impact of wind turbines on property values. In- stead, the literature strongly suggests that the specic context of each facility will drive outcomes for that community. This paper takes the next step of trying to determine what some factors might be by examining for impacts in the border region where part of our sample is closer to the facility, but also receives compensatory payments from the developer, while the rest of our sample views the turbines from a distance and does not receive any compensation.

3 Methodology

Hedonic analysis is a well-accepted form of revealed preference (as opposed to stated preference) non-market valuation in the eld of environmental economics. This approach is derived from Rosen [1974] and others.6 Essentially, Rosen [1974] lays out a model of buyers and sellers with preferences over the attributes of a compound good, like housing. In this case, he derives a hedonic price function showing that, under certain assumptions, the market price of a house will be a function of its attributes. One can then estimate this hedonic function using data on a set of homes with varying prices and attributes, where the estimated coecients represent the marginal willingness-to-pay for changes in these individual attributes. The strength of this approach is that it allows for the estimation of the value of marginal changes in attributes that are otherwise not bought and sold on markets, such as environmental amenities. The hedonic method is quite powerful because any amenity which is valued by consumers and associated in some way with housing markets can, in theory, be valued through this method under the right conditions. A limitation of this approach, however, is that the hedonic method can only estimate what are

6See III et al. [2014] and Taylor [2003]for comprehensive treatments of the hedonic method in environmental economics.

7 called use values of homeowners or renters, and not broader societal values. For instance, the hedonic method could estimate the value of having a healthy ecosystem to residents in a local area, but cannot estimate the value to society- at-large of a particular ecosystem. As in any econometric exercise, there are a number of empirical issues that are common in hedonic analyses. First, one must take care in choosing the functional form. Traditionally, based on Cropper et al. [1988], the log-linear or log-log forms are preferred. However, more recently, Kumino et al. [2010] advocate for testing the appropriateness of this functional form using the Box- Cox formulation (Box and Cox, 1964). Subsequent applications, including recent studies on the impacts of wind turbines on property values (Heintzelman and Tuttle [2012], Vyn and Mccullough [2014]), have found that the Box-Cox analysis supports the log-linear form in this context. Similarly, our robustness testing in this study indicates that the results based on a Box-Cox functional form are consistent with those of the log-linear form. As a result, we primarily use the log-linear form in our analysis. Kumino et al. [2010] also advocate for the inclusion of spatial xed eects, temporal controls, and quasi-experimental identication to control as best as possible for omitted variables bias, which is endemic to applications of the hedonic method. Omitted variables bias arises when one or more factors that are correlated with both the dependent variable and one or more included explanatory variables are omitted from the specication. In this case, the analysis will assign explanatory power which properly belongs with the omitted variable to included variables, resulting in biased estimates of the eect of those variables. Given the large number of factors which are both unobservable to the analyst and correlated with property values (local neighborhood attributes, for instance), this has serious implications for hedonic analysis. Similarly, given that we are trying to identify the impact of a feature which changes over time (the Wolfe Island turbines are built mid-sample period), if we fail to adequately control for trends over time, we may similarly bias our estimates of the turbine impacts. Fixed eects approaches help overcome these issues. By implicitly including a large number of spatial dummy variables, xed eects allows each area (township or census block, for instance) to have its own intercept term, accounting for time invariant factors which aect property values in these areas. The smaller the level of the xed eects, the less likely one is to have an omitted variables problem (as more and more spatial factors will be subsumed in the xed eect). However, xed eects also rely on within unit variation to identify the eects of remaining variables. This often means that as the level of the xed eects gets smaller, the analyst loses power to identify eects. Thus, it is important to carefully balance these eects when interpreting results and choosing a level of xed eects. In our case, we found that the use of xed eects at the township level provided an appropriate balance, as our preliminary analysis indicated that the ability to identify eects was diminished with the use of xed eects at the census block level. In addition, dierences exist in the specication of the American census blocks and the closest Canadian equivalent, the homogeneous

8 neighborhood which would undermine our ability to compare results across the border. We include year xed eects to account for sample wide trends as well as month xed eects to account for seasonality in prices. This is particularly important given the turbulent real estate markets of the late 2000s.7 Our quasi- experimental identication stems from the fact that we have transactions taking place both before and after the turbines were built and in areas that are at varying distances and with varying views of the turbines. We include clustered error terms at the same level as our xed eects. This allows error terms to be correlated across transactions in the same local area (i.e., townships) while requiring them to be independent across local areas. This generalization helps to control for spatial autocorrelation and is a simplied form of spatial econometrics. With all of this in mind, our specication follows Heintzelman and Tuttle [2012] and takes the general form:

ln(Pijt) = λt + αj + β1windi + β2windt + β3windit + β4xit + ηjt + ijt (1) where λt represents the time (year and month) xed eects, αj represents local area xed eects, windi represents the proximity to or view of wind turbines for parcel i, windt represents the time period, t, in which turbine impacts are expected to occur, windit represents the interaction term between proximity to or view of wind turbines for parcel i with time period t, xit is a vector of other parcel and structural characteristics, and ηjt and ijt represent spatial area and individual specic error terms. The model represented in equation (1) is estimated separately for real estate markets in Ontario and New York, which permits determining whether the turbine impacts diered between the two locations.

4 Data

We use data on 8,279 single-family residential property transactions on both sides of the border in our analysis: 6,017 in Jeerson County, NY8, and 2,262 across the border in Frontenac County, Ontario. 1 provides a map of the study area and transaction locations. Data on NY transactions between January 2004 and July 2013, inclusive, come from the New York State Oce of Real Property Taxation Services (NYSORPTS). This data includes sale price, sale date, and parcel identiying information. This transaction data is then merged with parcel and home characteristics data from the asseessment process, also from NYSORPTS. We bring in parcel shapele (GIS) data which we acquired

7It is important to note that real estate markets in Northern New York and in Ontario were relatively insulated from this turmoil, and there is no dramatic crash in prices following the nancial crisis in our sample. Nonetheless, it is important to control for these possible impacts. 8There were an additional 11 transactions that had to be omitted from the dataset due to incomplete information.

9 from the Jeerson County Assessor's Oce. With this spatial data we calcu- late a number of distance and spatial variables in ArcGIS. Data on Canadian transactions comes from Ontario's Municipal Property Assessment Corporation (MPAC). This detailed data includes all open-market sales of residential properties in Frontenac County between September 2004 and July 2013, inclusive. An extensive set of property and structural variables is included in the MPAC data, while additional distance and spatial variables are calculated using Ar- cGIS. There are no sales in this dataset of properties on which a turbine is located. The key variables that capture the visual and aural impacts of the wind turbines include both visibility and distance variables, and we estimate separate models using each of these variables. The distance from each parcel to the nearest turbine on Wolfe Island was calculated using ArcGIS. Previous studies that have accounted for turbine impacts on property values based on distance measures have used inverse distance (e.g., Heintzelman and Tuttle [2012],Vyn and Mccullough [2014]) and distance bands (e.g.,Hoen et al., 2011, 2015). Both of these measures have potential shortcomings: the use of inverse distance involves estimating impacts at the mean distance from the turbines and, as such, may underestimate the potentially greater impacts in very close proximity to turbines, while the use of distance bands, which involves dividing up observations into bins of specied distances from turbines, can result in estimated impacts for each band that may be based on relatively few observations. Due to the relatively low number of observations in our datasets in close proximity to the turbines, the shortcoming associated with the use of distance bands is of greater concern. As a result, we use the inverse distance measure in our analysis. Given the distance-decaying nature of the visual and aural impacts of turbines, any impacts on property values are expected to be greater in closer proximity to the turbines, which would be reected by a negative coecient for the inverse distance variable. Since the data includes sales up to 71 miles from the turbines on the Canadian side and up to 44 miles away on the American side, there are a considerable number of sales that would not be impacted by the turbines, which comprise a control group. Since proximity does not necessarily correspond with visual impact, as view may be obstructed by landscape features, we also conducted eld visits to all parcels that were potentially within visibility of the wind turbines (this potential was determined based on a combination of viewshed modelling in ArcGIS and distance to the nearest turbine). On these visits, it was determined whether the parcel (as viewed from the road) had a partial or full view of one or more of the turbines. The eld observations of the visual impacts of every transacted parcel with a potential view is a strength of this paper, similar to that of Hoen et al. [2011] and Vyn and Mccullough [2014]. It is superior to relying solely on viewshed analysis in ArcGIS, which generally relies on a number of assumptions about the height of dierent forms of land cover (trees) to estimate views, and cannot possibly include every potential obstruction. Because of this, viewshed analysis will tend to overestimate turbine views, and we avoid this potential error with the eld observations.

10 Figure 1: Study Area

11 Since the data includes sales that occurred both before and after the wind farm was developed, it is necessary to account for the time period during which turbine impacts on property values are expected to occur. We specify a post- turbine period (Post-Con) to account for sales that occurred after turbine construction on Wolfe Island was completed in June 2009. In addition, we account for potential announcement eects by specifying an announcement period between April 2007, when the wind farm was rst announced, and June 2009 (Announcement). Each of these time period variables is interacted with the visibility (View) and turbine distance (InvDistance) variables in order to specify the variables that account for the impacts of turbines on property values. How- ever, given the diculty in specifying the precise point in time at which impacts may begin to occur, we test the robustness of our results with alternate specications of these time periods. Rather than using the post-construction date for the specication of the post-turbine period, we use the date that construction began: May 2008. During the construction period, while all turbines were not yet fully erected, the locations were known and the visual eects were becoming evident; hence, turbine impacts could conceivably have begun to occur during this period. We also use an alternate announcement eect period that begins following the approval of the zoning bylaw amendments in November 2006. In the post-construction period, the Canadian data includes 47 parcels within ve miles of the nearest turbine and 19 parcels within one mile, while the Amer- ican data includes 58 parcels within ve miles, the nearest of which is 1.86 miles from the turbines. There are 39 parcels in Ontario and 15 parcels in New York with a view of the turbines that were sold in the post-construction period. Since these numbers of observations are relatively low, we do not break down these parcels between those that have a full view and those with a partial view; rather, we only use the measure of whether a parcel has any view of a turbine. Any eects that we estimate using this measure will be an average of eects across these sub-categories of view, presumably over-estimating the eect for those with only a partial view of one turbine and under-estimating the eect for those with full views of multiple turbines. To ensure consistency in the estimation approach between the two sides of the border and to reduce the possbility of bias between the two sets of results, the same set of explanatory variables representing the parcel and structural characteristics are used. Variables accounting for parcel attributes include lot size and categorical variables for waterfront and seasonal properties. The value associated with the residence on each parcel is accounted for by a set of variables that includes the numbers of bathrooms, bedrooms, and stories, the living area, age of the house, and categorical variables for the existence of a replace, central air conditioning, and a basement, quality ratings (1-5 on a 5-point scale), heat types (electric, forced air, hot water, other, and no heat), and whether the residence is a mobile home. In addition to distance to the nearest turbine, other distance variables include the distances from each parcel to the nearest city and town. As mentioned above, sets of year and month xed eect variables are also included. Most of these variables were included in both datasets, though in some cases slight adjustments were made in the merging process. For example,

12 the variable representing house quality is based on a 5-point scale for the NY data and on a 10-point scale for the ON data, which is accounted for by dividing the quality index by two for parcels in Ontario. Summary statistics for all variables for the two datasets are presented in Table 1. Notice that property values in our Ontario sample are considerably larger than in our New York sample. A higher proportion of homes in our NY sample are on the waterfront as compared to our ON sample, and a higher proportion are also seasonal in our NY sample. Together, these facts suggest that a higher proportion of our NY sample, and particularly those properties aected by the turbines, are likely to be vacation or second homes. Otherwise, our samples are quite comparable along most dimensions.

5 Results

Table 2 presents the results of our analysis for the regressions in which turbine impacts are accounted for by visibility, while Table 3 presents the results for the regressions based on distance to the nearest turbine. For the sake of brevity, only the results for the variables on which we focus our discussion, which include the turbine variables as well as the waterfront and seasonal variables, are reported in these tables.9

In Table 2, the key variables of interest are the interaction terms that indicate how turbine view impacts property values in the announcement and post-construction periods (View*Announcement; View*Post-Con). First, we nd negative and signicant impacts from turbine construction for homes in NY with a turbine view. The estimated coecient for this variable indicates that the values of homes in NY with a view of the turbines have, on average, been reduced by 13.4% following construction of the turbines10. Conversely, no signicant impact is observed on the Canadian side in either period. It is also evident from the results that sale prices across the NY sample decreased in the post-construction period relative to prices prior to this period, but this decrease was signicantly greater for parcels that had a view of the turbines. A positive impact is found in NY during the announcement period, but conceivably during this time home buyers in New York may not have been aware of the proposed wind farm or the location of the turbines to be constructed, or at the very least may not have been aware of the impending impact on their viewshed. This positive impact may also be the result of an omitted variable related to water view on the NY side. The property type has a major impact on sale price in both NY and ON. Waterfront properties sell at a substantial premium on both sides of the border,

9Estimated coecients for other variables, which include parcel and structural variables (size of home, etc.) as well as xed eects variables, are largely consistent with expectations. These results are available from the authors upon request. 10This gure is derived based on coecient interpretation for categorical variables in semi- log equations (see Halvorsen and Palmquist, 1980)

13 Table 1: Summary Statistics Ontario New York Variable Name Variable Description Mean Std. Dev. Min Max Mean Std. Dev. Min Max

Sale Price Sale price of the parcel, in Canadian $ for 241,150.40 122,104.90 25,000.00 1,600,000.00 142,004.90 101,407.80 3,500.00 2,000,300.00 Ontario sales and in US $ for New York sales View = 1 if parcel has a view (full or partial) of the 0.034 0.18 0.00 1.00 0.004 0.07 0.00 1.00 turbines Distance Distance to Nearest Wolfe Island Turbine, in 19.28 13.21 0.28 71.45 25.75 9.02 1.86 44.38 miles Announcement = 1 if parcel sold between project 0.354 0.48 0.00 1.00 0.211 0.41 0.00 1.00 announcement and completion of turbine construction Post-Con = 1 if parcel sold following completion of 0.586 0.49 0.00 1.00 0.410 0.49 0.00 1.00 turbine construction Waterfront = 1 if parcel is on the waterfront 0.049 0.22 0.00 1.00 0.117 0.32 0.00 1.00 Seasonal = 1 if parcel is seasonal 0.064 0.24 0.00 1.00 0.102 0.30 0.00 1.00 Mobile = 1 if residence is mobile home 0.012 0.11 0.00 1.00 0.037 0.19 0.00 1.00 Lot Size Size of the parcel, in acres 4.184 14.48 0.06 300.56 6.538 25.41 0.01 391.43 Bathrooms Number of bathrooms 1.515 0.70 0.00 4.50 1.433 0.58 0.00 5.50

14 Bedrooms Number of bedrooms 2.924 0.82 0.00 9.00 2.990 0.94 0.00 9.00 Fireplace = 1 if at least one replace exists in the house 0.286 0.45 0.00 1.00 0.170 0.38 0.00 1.00 Air = 1 if house has central air conditioning 0.234 0.42 0.00 1.00 0.021 0.14 0.00 1.00 Quality1 = 1 if house quality index is 1 0.010 0.10 0.00 1.00 0.017 0.13 0.00 1.00 Quality2 = 1 if house quality index is 2 0.061 0.24 0.00 1.00 0.124 0.33 0.00 1.00 Quality3 = 1 if house quality index is 3 0.866 0.34 0.00 1.00 0.763 0.43 0.00 1.00 Quality4 = 1 if house quality index is 4 0.063 0.24 0.00 1.00 0.095 0.29 0.00 1.00 Quality5 = 1 if house quality index is 5 0.000 0.00 0.00 0.00 0.001 0.03 0.00 1.00 Living Area Living area of the house, in square feet 1493.10 587.71 101.00 4839.00 1,557.52 598.32 136.00 6,074.00 Age Age of the house 42.45 37.29 0.00 190.00 67.50 52.45 0.00 225.00 Stories Number of stories in the house 1.26 0.40 1.00 3.00 1.46 0.44 1.00 3.00 Electric = 1 if the house has electric heat 0.18 0.38 0.00 1.00 0.15 0.35 0.00 1.00 Forced Air = 1 if the house has forced air heat 0.72 0.45 0.00 1.00 0.70 0.46 0.00 1.00 Hot Water = 1 if the house has hot water heat 0.02 0.14 0.00 1.00 0.08 0.27 0.00 1.00 Other = 1 if the house has other heat type 0.04 0.20 0.00 1.00 0.00 0.00 0.00 0.00 No Heat = 1 if the house has no heat 0.046 0.21 0.00 1.00 0.08 0.27 0.00 1.00 Basement = 1 if the house has a basement 0.33 0.47 0.00 1.00 0.09 0.28 0.00 1.00 Town Distance to the nearest town, in miles 13.80 10.29 0.00 57.02 1.61 1.51 0.00 8.25 City Distance to the nearest city (population > 17.02 16.49 0.00 81.32 2.90 3.03 0.00 17.88 50,000), in miles Observations 2,262 6,017 Table 2: Regression Results - Turbine View Ontario New York Variable coef se coef se View 0.008 0.017 0.045 0.048 Announcement -0.010 0.021 -0.100* 0.059 Post-Con 0.043 0.046 -0.174** 0.069 View*Announcement 0.015 0.048 0.207*** 0.056 View*Post-Con -0.004 0.040 -0.144*** 0.050 Waterfront 0.632*** 0.026 0.646*** 0.071 Seasonal -0.093*** 0.020 0.128*** 0.045 Number of observations 2,262 6,017 Adjusted R2 0.711 0.423

Table 3: Regression Results - Turbine Distance Ontario New York Variable coef se coef se InvDistance -0.189*** 0.027 -0.661* 0.369 Announcement -0.012 0.024 -0.104 0.065 Post-Con 0.034 0.049 -0.145* 0.074 InvDistance*Announcement 0.043 0.045 0.126 0.319 InvDistance*Post-Con 0.072* 0.039 -0.506** 0.242 Waterfront 0.559*** 0.019 0.648*** 0.069 Seasonal -0.086*** 0.018 0.123*** 0.047 Number of observations 2,262 6,017 Adjusted R2 0.716 0.424

15 while seasonal homes sell for a premium in the NY sample but at a discount in Ontario.11 The negative coecient for ON may be due to the prevalence of more rustic seasonal homes in the northern, mainland area of Frontenac County, which are used more for hunting and shing purposes than as vacation homes. In general, this tends to be the case for seasonal homes across NY, but in this study area seasonal homes are more likely to be vacation homes located on particularly attractive parcels, or in attractive locations. An important caveat about the post-construction impact observed in NY is that, while it is statistically signicant, it is based on a relatively small number of parcels (15) with a view of the turbines that were sold after the turbines were built. This suggests that this result should be interpreted cautiously. The statistical signicance likely stems from the large number of control transactions to which these transactions are being compared. The small number of treatment transactions is unfortunately common in hedonic studies of wind turbines. The results of the specication in which turbine impacts are accounted for by distance to the nearest turbine are reported in Table 3. This specication is less susceptible to the small numbers problem highlighted above since every parcel has a distance to the nearest turbine and as such can contribute to the estimation of turbine impacts. The key variables of interest are the interaction terms between the inverse of the distance to the nearest turbine and each of the announcement and post-construction periods (InvDistance*Announcement; InvDistance*Post-Con). We nd that the results of this specication are broadly consistent with those based on turbine view. In particular, we nd, again, a signicant negative impact in NY in the post-construction period, where parcels in closer proximity to turbines have experienced reductions in value relative to those at greater distances from the turbines. A positive impact is observed in the ON sample in the post-construction period, which is unexpected, though it is only signicant at the 10% level. Inverse distance to turbine is negative and signicant in both samples, possibly lling in for a measure of proximity to the water, particularly in the NY sample. Seasonal and waterfront homes have the same eects as seen above. Overall, the results of these two specications indicate that negative impacts associated with the Wolfe Island turbines have only occurred to any observable extent on the NY side, but not in Ontario despite the fact that many parcels on Wolfe Island are located in amongst the turbines.This is somewhat surprising given that concerns were raised on both sides of the border, but may reect the cross-border dierence in the level of involvement in the project development process. In addition, as indicated in Fast et al (2015), there are a fair number of residents, both long-time residents and newcomers, on Wolfe Island that are quite supportive of the wind farm. Those who support the wind farm are less likely to be concerned about impacts on property values and may not reduce their willingness-to-pay for properties with a view of or in close proximity to the turbines. If a large enough proportion of residents and potential homebuyers

11NYSORPTS denes seasonal homes as those Dwelling units generally used for seasonal occupancy; not constructed for year-round occupancy (inadequate insulation, heating, etc.). MPAC is less specic about this denition for seasonal homes.

16 support the wind farm and believe it benets the community, this may reduce the likelihood or magnitude of impacts on property values. It is also true that the township of Frontenac Islands receives an annual payment of C$650,000 from the developer that should also be working counter to negative impacts. Due to the uncertainty around the time at which turbine impacts could conceivably begin to occur, we test an alternate specication of the post-turbine period. In this specication we use the date that turbine construction began (May 2008) as the date following which impacts would be expected to occur. As evident in Tables 4 and 5, the results are consistent with those of the post-construction specication. We also test an alternate announcement period where, as described above, we specify this period to begin following approval of zoning bylaw amendments rather than using the date of announcement. The results of this specication are again similar to those of our base specication, but for the distance model the negative impact in the post-construction period for the NY sample is not statistically signicant.12 One of the underlying assumptions of the hedonic method is that the sample area represents a single market. However, in our study areas this assumption may not necessarily hold due to dierences in the composition of residential property types and usage across each area. For example, on both sides of the border, the areas that would be most aected by the wind turbines (i.e., Wolfe Island and Cape Vincent), have proportionally greater waterfront properties relative to the other areas of their respective counties, and are also used to a greater extent for vacation purposes. Such dierences could potentially introduce bias into our estimates of the turbine impacts. To address this potential bias, we consider a specication that restricts our samples to parcels within 5 miles of the turbines.13 The results for the NY sample for both the view and distance specications indicate similar results as with the full sample, where signicant negative impacts are found in the post-construction period. Conversely, signif- icant positive impacts are found for the ON sample in the post-construction period, which coincides with the full sample nding for the distance specication but not with the view specication. However, as with our full sample specications, a negative impact is not found for the ON sample.

6 Conclusions

The main contributions of this paper are that it provides evidence that impacts of wind turbines on property values vary depending on the context and it iden- ties some factors that may contribute to the occurrence of signicant impacts. This paper adds to the growing amount of evidence that wind facilities do have signicant economic impacts on local communities. In our particular context, with a unique scenario in which two communities separated by an international

12We nd similar results when we combine the alternate specications for both the announcement and post-turbine periods. 13We also tested a 10-mile restriction and found similar results to that of the 5-mile restriction.

17 Table 4: Results for Robustness Checks - Turbine View Pre-construction Announcement Within 5 Miles of Turbines Ontario New York Ontario New York Ontario New York Variable coef se coef se coef se coef se coef se coef se View 0.003 0.017 0.047 0.048 -0.040** 0.016 0.044 0.048 -0.150*** 0.012 0.706** 0.282 18 Announcement -0.021 0.018 -0.093 0.060 -0.249*** 0.069 -0.017 0.085 -0.225*** 0.066 -0.336 0.368 Post-Turbine -0.061 0.043 -0.074 0.083 -0.194*** 0.074 -0.085 0.103 -0.357*** 0.116 0.020 0.661 View*Announcement 0.031 0.055 0.210*** 0.060 0.118*** 0.044 0.196*** 0.054 0.271*** 0.053 -0.308 0.401 View*Post-Turbine 0.002 0.040 -0.146*** 0.049 0.054 0.039 -0.142*** 0.049 0.258*** 0.064 -0.797** 0.360 Number of observations 2,262 6,017 2,262 6,017 110 129 Adjusted R2 0.711 0.423 0.711 0.423 0.774 0.551 Table 5: Results for Robustness Checks - Turbine Distance Pre-construction Announcement Within 5 Miles of Turbines Ontario New York Ontario New York Ontario New York Variable coef se coef se coef se coef se coef se coef se InvDistance -0.191*** 0.027 -0.669* 0.368 -0.183*** 0.024 -0.829** 0.399 -0.178*** 0.001 2.139 2.511 19 Announcement -0.021 0.021 -0.111* 0.067 -0.131 0.080 -0.047 0.079 -0.269*** 0.010 -0.341 1.037 Post-Turbine -0.063 0.041 -0.045 0.084 -0.085 0.092 -0.074 0.100 -0.514*** 0.070 2.284** 1.092 InvDistance*Announcement 0.034 0.051 0.395 0.321 0.032 0.037 0.446 0.345 0.128*** 0.014 0.932 2.723 InvDistance*Post-Turbine 0.070* 0.039 -0.549** 0.242 0.066* 0.035 -0.357 0.271 0.198*** 0.038 -4.706* 2.578 Number of observations 2,262 6,017 2,262 6,017 110 129 Adjusted R2 0.716 0.424 0.716 0.424 0.781 0.550 border are aected by the same wind farm, we nd evidence that property values are negatively impacted by wind turbines, but only in our NY sample, which is at a greater distance from the turbines. However, many of the aected homes in that sample are likely to be vacation homes (we, unfortunately, cannot posi- tively identify which ones) and homes with a view of the turbines over the water - a view that may have been a main amenity of those properties when last transacted. In addition, these American homeowners had no ability to participate in the siting process, nor do they recieve any compensation from the development. Hence, these may be some contextual factors that determine whether property values in a local area are negatively impacted by wind turbines. There are limitations in this study that should be acknowledged. First, as noted by Hoen et al. [2015], the use of inverse distance to account for turbine impacts estimates this impact at a mean distance from the turbines, which can hamper the ability to generate accurate impacts in close proximity to the turbines. This issue could potentially be avoided through the use of discrete distance bands. However, this approach may be unable to detect signicant impacts within specied bands if the number of aected sales is relatively low, which is the case in our study. Interestingly, the nding of signicant negative impacts on property values on only the U.S. side occurred despite the fact that there were fewer sales with a view of the turbines than on the Ontario side. This suggests that the lack of evidence of negative impacts on the Ontario side is not necessarily due to a lack of observations. However, the possibility remains that parcels that are impacted the most by the turbines were not sold or were unable to be sold. Unfortunately, data is not available to examine this issue; as such, this represents an important caveat to our study results. This study furthers the conclusion from a read of the literature on wind turbines that the specic context of each facility has the potential to make a large dierence in the nature of the economic impacts. In our case, we have impacts that dier across two communities impacted by the same facility. This implies that, in general, researchers should not expect there to be one single answer to the question of how wind farms aect property values, and that the lack of consensus in the literature is not necessarily problematic. Instead, making any forecast of impacts will require a more careful comparison to communities that have already been studied. As in any benets transfer process, nding a proper comparison site is the critical task when using results from one community to predict outcomes for another.

20 References

Jamie Baxter, Rakhee Morzaria, and Rachel Hirsch. A case-control study of support/opposition to wind turbines: Perceptions of health risk, economic benets, and community conict. Energy Policy, 61:931943, October 2013. ISSN 0301-4215. doi: 10.1016/j.enpol.2013.06.050. Karl Bolin, Gösta Bluhm, Gabriella Eriksson, and Mats E. Nilsson. Infrasound and low frequency noise from wind turbines: exposure and health eects. Environmental Research Letters, 6(3):035103, 2011. ISSN 1748-9326. doi: 10.1088/1748-9326/6/3/035103. G. E. P. Box and D. R. Cox. An Analysis of Transformations. Journal of the Royal Statistical Society. Series B (Methodological), 26(2):211252, 1964. ISSN 0035-9246. Jason P. Brown, John Pender, Ryan Wiser, Eric Lantz, and Ben Hoen. Ex post analysis of economic impacts from wind power development in U.S. counties. Energy Economics, 34(6):17431754, November 2012. ISSN 0140-9883. doi: 10.1016/j.eneco.2012.07.010. Council of Canadian Academies, Expert Panel on Wind Turbine Noise and Human Health. Understanding the evidence: wind turbine noise. 2015. ISBN 978-1-926522-07-4. Maureen L. Cropper, Leland B. Deck, and Kenenth E. McConnell. On the Choice of Functional Form for Hedonic Price Functions. The Review of Economics and Statistics, 70(4):668675, 1988. ISSN 0034-6535. doi: 10.2307/1935831.

Stewart Fast, Warren Mabee, and Jean Blair. The changing cultural and economic values of wind energy landscapes. The Canadian Geographer / Le Géographe canadien, 59(2):181193, June 2015. ISSN 1541-0064. doi: 10.1111/cag.12145. Stephen Gibbons. Gone with the wind: Valuing the visual impacts of wind turbines through house prices. Journal of Environmental Eco- nomics and Management, 72:177196, July 2015. ISSN 0095-0696. doi: 10.1016/j.jeem.2015.04.006. Theresa M. Groth and Christine A. Vogt. Rural wind farm development: Social, environmental and economic features important to local residents. Renewable Energy, 63:18, March 2014. ISSN 0960-1481. doi: 10.1016/j.renene.2013.08.035. Robert Halvorsen and Raymond Palmquist. The Interpretation of Dummy Vari- ables in Semilogarithmic Equations. American Economic Review, 70(3):474 75, 1980.

21 Martin D. Heintzelman and Carrie M. Tuttle. Values in the Wind: A Hedonic Analysis of Wind Power Facilities. Land Economics, 88(3):571588, August 2012. ISSN 0023-7639, 1543-8325. doi: 10.3368/le.88.3.571. Ben Hoen, Ryan Wiser, Peter Cappers, Mark Thayer, and Gautam Sethi. Wind Energy Facilities and Residential Properties: The Eect of Proximity and View on Sales Prices. Journal of Real Estate Research, 33(3):279316, January 2011. ISSN 0896-5803. doi: 10.5555/rees.33.3.16133472w8338613. Ben Hoen, Jason P. Brown, Thomas Jackson, Mark A. Thayer, Ryan Wiser, and Peter Cappers. Spatial Hedonic Analysis of the Eects of US Wind Energy Facilities on Surrounding Property Values. Journal of Real Estate Finance and Economics, 51(1):2251, July 2015. ISSN 0895-5638. doi: 10.1007/s11146- 014-9477-9. WOS:000355662300002. A. Myrick Freeman III, Joseph A. Herriges, and Catherine L. Kling. The Mea- surement of Environmental and Resource Values: Theory and Methods. Rout- ledge, June 2014. ISBN 978-1-317-70392-1.

Cathrine Ulla Jensen, Toke Emil Panduro, and Thomas Hedemark Lundhede. The Vindication of Don Quixote: The Impact of Noise and Visual Pollution from Wind Turbines. Land Economics, 90(4):668682, November 2014. ISSN 0023-7639, 1543-8325. doi: 10.3368/le.90.4.668. Daniel T. Kane, Brannin J. McBee, and Jozef Lieskovsky. Emissions savings from wind power generation in Texas. The Energy Journal, 34(1):155, 2013. Matthew E. Kahn. Local non-market quality of life dynamics in new wind farms communities. Energy Policy, 59:800807, August 2013. ISSN 0301-4215. doi: 10.1016/j.enpol.2013.04.037.

R. Keating. Wolfe Island is proposed as future home to 86 windmills. Watertown Daily Times, January 2006. Nicolai V. Kumino, Christopher F. Parmeter, and Jaren C. Pope. Which hedonic models can we trust to recover the marginal willingness to pay for environmental amenities? Journal of Environmental Economics and Management, 60(3):145160, November 2010. ISSN 0095-0696. doi: 10.1016/j.jeem.2010.06.001. Jacob Ladenburg, Mette Termansen, and Berit Hasler. Assessing acceptability of two onshore wind power development schemes: A test of viewshed eects and the cumulative eects of wind turbines. Energy, 54:4554, June 2013. ISSN 0360-5442. doi: 10.1016/j.energy.2013.02.021.

Corey Lang, James J. Opaluch, and George Snarolakis. The windy city: Prop- erty value impacts of wind turbines in an urban setting. Energy Economics, 44:413421, July 2014. ISSN 0140-9883. doi: 10.1016/j.eneco.2014.05.010.

22 Steven Laposa and Andrew Mueller. Wind Farm Announcements and Rural Home Prices: Maxwell Ranch and Rural Northern Colorado. Journal of Sustainable Real Estate, 2(1):383402, January 2010. ISSN 1949-8276. doi: 10.5555/jsre.2.1.b3lt843187581v40. Andreas Lindén, Lauri Rapeli, and Andreas Brutemark. Community attach- ment and municipal economy: Public attitudes towards wind power in a local context. Environmental Science & Policy, 54:1014, December 2015. ISSN 1462-9011. doi: 10.1016/j.envsci.2015.06.005. M. Lynne. A Blight On Landscape. Watertown Daily Times, June 2009. M. Munday, G. Bristow, and R. Cowell. Wind farms in rural areas: How far do community benets from wind farms represent a local economic development opportunity? Journal of Rural Studies, 27(1):112, 2011. ISSN 0743-0167. doi: 10.1016/j.jrurstud.2010.08.003. Ontario Power Authority. Wolfe Island Wind Project (197.8 MW) - Wolfe Island. Maria A. Petrova. NIMBYism revisited: public acceptance of wind energy in the United States. Wiley Interdisciplinary Reviews-Climate Change, 4 (6):575601, November 2013. ISSN 1757-7780. doi: 10.1002/wcc.250. WOS:000325619500008. Sidney Quarrier. Turbine Strobe Lights Dominate Landscape. Watertown Daily Times, September 2009. Jerry Radley. Beauty of Cape Vincent Area Is Threatened. Watertown Daily Times, May 2009. Sherwin Rosen. Hedonic prices and implicit markets: product dierentiation in pure competition. The journal of political economy, pages 3455, 1974. Chris Shultz, Joshua C. Hall, and Michael P. Strager. Production of Wind Energy and Agricultural Land Values: Evidence from Pennsylvania. Working Paper 15-11, Department of Economics, West Virginia University, 2015. S. Sims and P. Dent. Property stigma: Wind farms are just the latest fashion. Journal of Property Investment and Finance, 25(6):626651, 2007. ISSN 1463- 578X. doi: 10.1108/14635780710829315. S. Sims, P. Dent, and G.R. Oskrochi. Modelling the impact of wind farms on house prices in the UK. International Journal of Strategic Property Management, 12(4):251269, 2008. ISSN 1648-715X. doi: 10.3846/1648- 715X.2008.12.251-269. Yasin Sunak and Reinhard Madlener. The Impact of Wind Farms on Property Values: A Geographically Weighted Hedonic Pricing Model. SSRN Scholarly Paper ID 2114216, Social Science Research Network, Rochester, NY, May 2012.

23 Laura O. Taylor. The Hedonic Method. In Patricia A. Champ, Kevin J. Boyle, and Thomas C. Brown, editors, A Primer on Nonmarket Valuation, number 3 in The Economics of Non-Market Goods and Resources, pages 331393. Springer Netherlands, 2003. ISBN 978-1-4020-1445-1 978-94-007-0826-6. DOI: 10.1007/978-94-007-0826-6_10.

R.J. Vyn and R.M. Mccullough. The Eects of wind turbines on property values in Ontario: Does public perception match empirical evidence? Canadian Journal of Agricultural Economics, 62(3):365392, 2014. ISSN 0008-3976. doi: 10.1111/cjag.12030. Jeremy Glenn Weber, Jason P. Brown, and John Pender. Rural Wealth Cre- ation and Emerging Energy Industries: Lease and Royalty Payments to Farm Households and Businesses. SSRN Scholarly Paper ID 2307667, Social Science Research Network, Rochester, NY, June 2013.