Abstract
The lack of consensus in the literature on wind turbines and property values suggests that factors such as attitudes toward wind energy may influence whether turbines will impact property values. Based on identified attitudes at the municipality level in the Canadian province of Ontario, this study compares impacts on property values, estimated using a difference-in-differences hedonic approach, between municipalities that oppose wind energy development and those that have not expressed opposition. The results indicate that wind turbines have negatively impacted property values in “unwilling host” municipalities, while no significant impacts are found in unopposed municipalities. (JEL R11, R52)
1. Introduction
The literature on property value impacts of wind turbines has grown substantially in recent years in response to the rapid expansion of wind energy development and the ensuing concerns expressed by local residents regarding potential impacts on property values, an issue that has become quite prominent and contentious in a number of jurisdictions. However, a general consensus on this issue has not been reached; while some studies have found evidence that wind turbines have negatively impacted surrounding property values (Heintzelman and Tuttle 2012; Jensen, Panduro, and Lundhede 2014; Gibbons 2015; Dröes and Koster 2016; Sunak and Madlener 2016), other studies have not found significant impacts on property values (Sims, Dent, and Oskrochi 2008; Hoen et al. 2011; Lang, Opaluch, and Sfinarolakis 2014; Vyn and McCullough 2014; Hoen et al. 2015).
The lack of consensus in the literature may have initially been attributable in part to the prevalence of studies in which impacts were estimated based on relatively few observations in close proximity to wind turbines. This issue may have inhibited the ability to detect significant impacts, particularly since impacts on property values are likely to be relatively small (Hoen et al. 2015). In fact, this issue was pervasive among early studies that did not find significant impacts. More recent studies have avoided this issue through the use of larger data sets, and have subsequently been more likely to find evidence of negative impacts (Gibbons 2015; Dröes and Koster 2016; Sunak and Madlener 2016). However, even among studies that avoid the small numbers issue, a lack of consensus still exists, as other recent studies with large numbers of observations in close proximity to turbines have not found significant impacts (Lang, Opaluch, and Sfinarolakis 2014; Hoen et al. 2015). This suggests that the lack of consensus is not solely due to data shortcomings; rather, there may be another explanation for the differences in results among these studies. For example, there may be contextual factors specific to each jurisdiction that influence whether wind turbines will impact property values, such as the level of involvement that residents have in the approval process for a wind energy facility, the number of turbines constructed, and the amount of compensation provided to the community by the developer. Indeed, Heintzelman, Vyn, and Guth (2017) speculated that such factors may have contributed to differences in impacts on property values observed in their study between two bordering jurisdictions from a single wind farm constructed in one of the jurisdictions.
The variation in the results across previous studies could also be attributed to heterogeneous attitudes toward wind energy (which could arise due to differences in contextual factors). A related body of literature has demonstrated evidence of differences across jurisdictions in wind energy attitudes and in perceptions of impacts, which may contribute to differences in observed impacts on property values. If residential property buyers support wind energy and do not perceive wind turbines to impact property values, they would not adjust their bids to reflect a disamenity impact. Conversely, buyers who perceive turbines to have a negative impact on property values would reduce their bids for affected properties, or would not bid at all. Accordingly, if a community has a relatively high proportion of buyers who perceive turbines to be a disamenity, the resulting lower demand for properties in closer proximity to turbines would cause a reduction in prices for these properties, while a community with a low proportion of buyers who perceive turbines to be a disamenity is unlikely to experience a similar decline in demand for these properties. Hence, this heterogeneity in perceptions may result in different levels of demand across jurisdictions for properties in close proximity to turbines, which may be contributing to the lack of consensus in the literature regarding property value impacts.
The possibility that variation in impacts could be due to differences in perceptions for wind energy has not been examined empirically. The primary impediment to examining this issue is difficulty in measuring and identifying differences in attitudes or perceptions across property buyers or across jurisdictions. In general, studies on attitudes toward wind energy are conducted through surveys of individuals, typically within a single jurisdiction, from which it can be difficult to extrapolate results to identify differences across jurisdictions.
However, the Canadian province of Ontario provides a setting that permits the identification of differences in attitudes toward wind energy at the municipality level. Some municipalities in Ontario have encouraged wind energy development, while other municipalities have fought to prevent the siting of wind farms within their communities. The ability to observe these municipal-level responses to wind energy development enables the evaluation and comparison of property value impacts between jurisdictions with different attitudes: that is, those that are opposed to wind energy development in their communities versus those that are not opposed.
This paper uses a difference-in-differences hedonic approach to estimate the impacts of wind turbines on rural residential property values in Ontario, and takes advantage of the identified attitudes toward wind energy to compare the estimated impacts between municipalities opposed to and municipalities unopposed to wind energy development. The results of this paper indicate differences in the estimated impacts of turbines on property values between the two sets of municipalities, where negative impacts are found in municipalities that have expressed opposition to wind energy development, while no significant impacts are found in unopposed municipalities. Hence, this paper addresses a gap in the literature by providing evidence that may explain the variation in results across previous related studies.
Literature on Wind Energy Attitudes and Perceptions
It is evident from the literature that heterogeneous attitudes toward and perceptions of wind energy exist. For example, Baxter, Morzaria, and Hirsch (2013) compared perceptions of wind energy between two Ontario communities—one with turbines and one without—and found significantly more support for wind energy in the community that already had turbines. This study also found that residents of the community with turbines were less concerned about impacts on health and property values. Direct relationships between proximity to turbines and support for wind energy have also been observed in other studies (Krohn and Dambourg 1999; Warren et al. 2005). However, evidence regarding the nature of these relationships is not consistent across all jurisdictions. For example, Swofford and Slattery (2010) found that support levels for wind energy in Texas were lower for residents in close proximity to turbines than for residents farther away, while other studies found no difference in support due to proximity (Johansson and Laike 2007; Graham, Stephenson, and Smith 2009). Hence, these results suggest that attitudes toward wind energy vary across jurisdictions, and are not necessarily linked to proximity of existing turbines.1
Attitudes toward wind energy can be linked to perceptions regarding potential impacts of turbines. Bidwell (2013) noted that attitudes toward wind energy may be developed in response to real or anticipated effects of wind farms. He also pointed out that while wind energy development can have both positive and negative impacts, undesirable effects have played a larger role in shaping public perception. In particular, attitudes toward wind energy are influenced to a large extent by perceived impacts related to the aesthetics of wind turbines (Warren et al. 2005). Another factor related to the influence of aesthetics is cumulative effects, where attitudes may also be influenced by the number of turbines in the viewshed. For example, Ladenburg and Dahlgaard (2012) and Ladenburg, Termansen, and Hasler (2013) found negative relationships between the number of turbines viewed daily and attitudes toward wind energy development. However, in a subsequent study, Ladenburg (2015) found that the number of turbines viewed daily had little to no impact on attitudes.
There are a number of other factors identified in the literature that can influence perceptions of wind energy and its associated impacts. Perceptions may be influenced by the opinions of people within one’s social network (Devine-Wright 2005), which implies that perceptions may vary across communities depending on the opinions and attitudes of a community’s most vocal or influential members. Similarly, perceptions may be influenced by the local media (Deignan, Harvey, and Hoffman-Goetz 2013). Perceptions of wind turbines may also be influenced by the level of involvement that local residents have in the development of a wind project (Heintzelman, Vyn, and Guth 2017).2
Differences across jurisdictions in perceptions of wind turbines and their potential impacts, which could be underlying the variation in the results of previous studies on property value impacts, have been observed in the literature. For example, Walker et al. (2014) found evidence that perceptions of the impacts of wind turbines on property values varied between residents of neighboring communities in Ontario, and suggested that social dynamics may have contributed to the differences in perceptions. It is noteworthy that the community in which residents were less likely to perceive negative impacts was located in close proximity to one of the first wind farms constructed in the province, while the community in which residents were more likely to perceive negative impacts was in close proximity to a more recently developed wind farm. This coincides with the findings of Warren et al. (2005), where perceptions of wind turbines were more negative among residents in close proximity to a proposed wind farm than among residents in close proximity to an existing wind farm. Van der Horst (2007) noted that this is consistent with the literature on risk perception, as anything new and unfamiliar can increase people’s dislike due to the higher level of perceived risk.
Negative perceptions regarding the impacts of wind turbines can contribute to opposition to wind energy development. Johansson and Laike (2007) found that people with negative attitudes toward wind energy, in particular due to the perceived visual effects, were more likely to have a strong intention to oppose turbines. A number of other factors can contribute to community opposition to wind energy projects, such as a perceived lack of fairness in the planning process (Gross 2007), lack of communication and consultation with residents (Krohn and Damborg 1999; Wolsink 2007), distrust of the developer, particularly for outside developers with no ties to the community (Jobert, Laborgne, and Mimler 2007), and the persuasiveness of local opposition groups (Van der Horst 2007).
Vocal opposition that focuses on the negative effects can in turn further contribute to shaping perceptions. Chapman, Joshi, and Fry (2014) examined this phenomenon in Australia, where they found that negative information presented by an anti–wind energy organization regarding health issues caused by wind farms impacted the expectations and perceptions of local residents. This suggests that the degree of negative attention wind energy receives could influence the perception of those who were previously indifferent to wind energy or who were not previously concerned about the potential impacts of wind turbines, which could result in greater homogeneity of attitudes (i.e., opposition to wind energy) for residents within a jurisdiction. This may have occurred to some degree in Ontario.
Background of the Study Area
The development of wind energy in the province of Ontario represents an interesting case study. When the first industrial wind farms were constructed in the mid-2000s, there was relatively little opposition to them and few concerns expressed publicly. These wind farms were sited in municipalities that chose to host them. A study on the first major industrial wind farm in the province found no significant impacts on property values (Vyn and McCullough 2014). The Green Energy Act, 2009, paved the way for the rapid expansion of the wind energy industry in the province. In an attempt to aggressively increase the proportion of energy derived from renewable sources, the provincial government offered significant financial incentives, primarily through energy rates much higher than the existing market rate, to encourage wind energy development.
The number of commercial wind turbines in the province increased from 10 in 2003 to over 2,300 in 2015, with the majority of this increase occurring after the implementation of the Green Energy Act. Figure 1 provides a map of the locations of wind farms in the province, which are primarily concentrated in rural areas along the Great Lakes. The rapid expansion of wind energy development captured the public’s attention, and many of the factors discussed above that influence wind energy attitudes and perceptions emerged during this period of expansion. While some public meetings were held to discuss proposed wind projects, local residents essentially had little to no influence regarding the approval and siting of these projects. Residents of many of the communities with recently constructed or proposed wind projects felt that the approval and siting processes were unfair, as the provincial government approved these projects despite feedback from residents, and even from municipalities, indicating their opposition to the projects and their concerns regarding impacts of wind turbines.
Map of Wind Farms in Southern Ontario Source: http://ontario-wind-turbines.org/.
The perceived unwillingness of the provincial government to listen to or address these concerns led to a groundswell of more vocal opposition and the formation of grassroots organizations, such as Wind Concerns Ontario, that actively opposed wind energy development on behalf of residents in communities across the province. Concerns from these opposition groups regarding impacts of wind turbines, such as impacts on health and property values, have been expressed prominently in local news media. This media attention, which has increased substantially in recent years, may have influenced attitudes toward wind energy and perceptions of turbine impacts.
Another source of controversy was the provincial government’s ability to override municipal decisions regarding wind energy development. To protest the lack of local control over the siting of wind farms, a large number of municipalities passed resolutions to declare their jurisdiction to be an “unwilling host” for wind energy (see Figure 2). While this designation could not prevent wind farms from being sited in their jurisdiction, it was a means to express their opposition to local wind energy development and to the province’s regulations regarding the siting of wind farms. While attitudes are unlikely to be uniform across all residents of unwilling host municipalities, it is assumed that the prevailing sentiment among residents of these municipalities would be opposition to wind energy.
Map of Unwilling Host Municipalities in Southern Ontario
Other municipalities, however, did not oppose wind energy development, and some actually encouraged it within their jurisdiction. For example, the municipality of Chatham-Kent encouraged the siting of wind projects within its borders, in part for economic reasons as the local economy had been suffering due to the substantial loss of manufacturing jobs. This coincides with a finding by Jobert, Laborgne, and Mimler (2007), where declining economic conditions made residents view a wind project proposal more favorably. This promotion of wind energy development within municipalities such as Chatham-Kent may indicate a difference in attitudes toward and perceptions of wind energy relative to unwilling host municipalities. The more favorable perceptions of wind energy may influence the nature of property value impacts observed in these communities. This paper examines the extent to which this has occurred in the province of Ontario.
The analysis conducted in this paper consists of three components. The first component involves estimating the average impacts of wind turbines on property values across southwestern Ontario, where wind energy development in the province has been concentrated. While an earlier study examined the impacts of the first industrial wind farm in the province and found no evidence of negative impacts (Vyn and McCullough 2014), an updated and more comprehensive study is needed to determine whether the rapid expansion of wind energy development in the province and the subsequent increase in concerns expressed publicly and through the media have contributed to a greater impact on property values. In the second component of the analysis, impacts are estimated and compared between the two groups of municipalities, based on the identified attitudes of municipalities toward wind energy development.
Finally, the third component of the analysis examines the impact of turbine density on property values. Previous studies have estimated density impacts for other disamenities such as oil and gas wells (Boxall, Chan, and McMillan 2005), shale gas wells (Gopalakrishnan and Klaiber 2014; Muehlenbachs, Spiller, and Timmins 2015), and livestock operations (Palmquist, Roka, and Vukina 1997; Herriges, Secchi, and Babcock 2005). In the literature on wind turbines and property values, which tends to focus on the impacts of turbine proximity or turbine visibility, the impact of turbine density has received relatively little examination, the results of which are varied. Gibbons (2015) found that the impact on property values increased with turbine density; conversely, Dröes and Koster (2016) found no evidence that additional turbines increased the impact on property values. Similarly, varied results are also found in stated preference studies regarding marginal willingness to pay for fewer turbines. For example, Meyerhoff, Ohl, and Hartje (2010) found that some segments of survey respondents were willing to pay for smaller wind farms, while other segments preferred wind farms with more turbines. The analysis conducted in the current study estimates the marginal effects of additional turbines within specific distance ranges and compares marginal effects between the two groups of municipalities.
2. Data
The data used to empirically evaluate the impact of wind turbines on property values in Ontario consists of rural residential property sales across southwestern Ontario between January 2002 and July 2013, as recorded by the Municipal Property Assessment Corporation (MPAC),3 the organization responsible for the assessment of all properties in the province. Observations that are not characterized as open-market sales are omitted from the data, as well as observations with missing information. In an effort to reduce the geographic extent of the study area and enhance the similarity of observations between the treatment and control groups, sales of properties greater than 20 km from the nearest wind turbine are excluded from the data. In addition, in order to avoid the influence of outliers, observations with the top 1% and bottom 1% of sale prices are omitted from the data.4 After these omissions, the data set used for the analysis consists of 22,159 sales. This data set includes a considerable number of variables that describe each property. A number of spatial variables, calculated using ArcGIS, are added to the data set, including the distance from each property to the nearest wind turbine as well as distances to urban areas.
Variables Accounting for Turbine Impacts
Appropriate specification of variables that account for the potential impact of wind turbines on property values is a relatively complicated endeavor, given the large number of wind farms in Ontario, many of which are in close proximity to each other and were constructed at different points in time. Thirty-seven wind farms were constructed in southwestern Ontario during the study period, the majority of which were completed between 2008 and 2012. These wind farms range considerably in size from 3 to 110 turbines. As evident in Figure 1, many wind farms are in close proximity to other wind farms. This complicates the specification of variables that account for turbine impacts, as in many cases the abutting wind farms were constructed at different points in time. Hence, this variable specification must account for the time period for each wind farm during which an impact is expected to occur.
While impacts on property values are assumed to be most likely to occur after the turbines have been constructed, an impact may also occur following the announcement of a wind farm, at which time the approximate location of the turbines may be known though the impact on the landscape (i.e., the disamenity) is not yet observable. In Ontario, the intense controversy surrounding wind energy development led to considerable media attention and public awareness of this issue, and public opposition to proposed projects began very early in the process. As a result, buyers of residential properties in areas with proposed wind farms would most likely have been aware of the impending potential disamenity.
Since the study data include sales that occurred both before and after the wind farms were announced and subsequently constructed, three time periods are specified for each wind farm: preannouncement (PA), announcement (AN), and postconstruction (PC). The preannouncement period occurs prior to the announcement of the wind farm, the announcement period covers the time period from the announcement of the wind farm to the construction of the turbines, and the postconstruction period is the time period following completion of turbine construction.
In addition to the temporal considerations, the variables accounting for turbine impacts must also take into account the relative proximity of properties to turbines. Based on the nature of the visual and aural disamenities associated with turbines, the impact is expected to be greatest in closest proximity to turbines and to diminish with distance from the turbines. To account for the variation in the expected impact that occurs with distance from turbines, previous studies have used either a continuous distance specification, such as inverse distance (e.g., Heintzelman and Tuttle 2012), or a discrete distance specification, such as through the use of distance bands (e.g., Hoen et al. 2015). While a continuous specification imposes structure on the data and may affect the accuracy of the estimated impact in close proximity to turbines, this approach is preferable in situations with limited sales data. Conversely, a discrete specification avoids the issues inherent in the continuous specification and permits the estimation ofimpacts within specific distance ranges. However, without adequate numbers of observations for each distance range, models based on this specification are unlikely to find statistically significant impacts unless the magnitude of the impact is quite large.
Since the data used for this analysis include a large number of sales in close proximity to turbines, the turbine impacts are captured through a set of distance bands, specified in 1 km increments up to 5 km, based on the distance to the nearest turbine. The numbers of sales in each distance band are provided in Table 1 for each time period. It is evident from this table that each band includes considerable numbers of sales, which supports the appropriateness of conducting the analysis using a discrete distance approach. By comparison, the study by Hoen et al. (2015) included 376 observations within 1 mi of turbines in the postconstruction period, while the current study includes 300 sales within 1 km (0.62 mi) in the postconstruction period.
Numbers of Sales within Specific Distance Ranges, by Time Period
The 5 km outer limit for expected impacts is based on a visual assessment of wind farms in Ontario, which indicated that visibility of turbines beyond this distance was negligible at best. This distance limit follows that of Vyn and McCullough (2014) and is similar to the 3 mi extent of anticipated impacts established by Hoen et al. (2015). To ensure that this distance limit is appropriate, additional specifications were tested to account for impacts beyond 5 km, but significant impacts were not found to occur beyond 5 km. As a result, all properties between 5 km and 20 km from the turbines are included in the control group and are not divided up into additional distance bands.
The distance band variables are then interacted with each of the time period variables to create the variables that account for the impact of turbines, where an interaction term is equal to 1 for a property for which the nearest turbine is within the corresponding distance band and time period, and 0 otherwise. Since impacts are anticipated to occur within 5 km of turbines and in both the announcement and postconstruction periods, the interaction terms for these time periods and distance bands are expected to capture the impacts of turbines.
However, as mentioned above, the specification of these variables is complicated by situations in which a property is in close proximity to multiple wind farms that were constructed at different times. For example, the sale of a property could occur during the announcement period for one wind farm in close proximity and during the postconstruction period for another. Of the 4,619 sales of properties within 5 km of at least one wind farm, 916 are within 5 km of two wind farms, 136 are within 5 km of three wind farms, 266 are within 5 km of four wind farms, and 94 are within 5 km of five wind farms. This implies that for some properties there could be multiple interaction terms categorized as 1, which raises the question as to how to appropriately account for turbine impacts for these properties; in other words, would the impact be equally attributable to all nearby turbines or just to the one that is most prominent?
Based on the assumption stated above that the potential impact associated with turbines declines with distance, this implies that the impact should be attributed primarily to the turbine from the closest wind farm rather than equally among all nearby wind farms.5 As such, for the subset of properties within 5 km of multiple wind farms in the announcement and postconstruction periods, only one turbine interaction term is categorized as 1, with all other interaction terms set equal to 0.
To determine the interaction term that captures the most prominent impact, and is thus categorized as 1, a ranking system is imposed based on two assumptions: (1) an impact is more likely to occur, or to be greater in magnitude, for the turbine in closest proximity; and (2) within each distance band an impact is more likely to occur, or to be greater in magnitude, in the postconstruction period than in the announcement period, due to the observability of the impact on the landscape. Hence, the ranking system is organized as follows: PC × 0–1 km, AN × 0–1 km, PC × 1–2 km, AN × 1–2 km, …, PC × 4–5 km, AN × 4–5 km, where a 1 is assigned to the highest-ranked interaction term for each property in close proximity to multiple wind farms.
The implications of this ranking system can be best explained through an example. For a property that is 1 to 2 km from one wind farm in the announcement period, 2 to 3 km from a second wind farm in the postconstruction period, and 4 to 5 km from a third wind farm in the postconstruction period, there would be three interaction terms that would initially be categorized as 1. In this case, based on the relative rankings, only AN × 1–2 km would be categorized as 1, while both PC × 2–3 km and PC × 4–5 km would be categorized as 0 rather than as 1.
However, the assumptions inherent in the imposed ranking system for specifying the interaction term that is set equal to 1 for properties close to multiple wind farms may influence the results. To address the potential bias that may arise for this subset of properties due to these assumptions, the robustness of the results under two alternate specifications of the interaction terms is examined.
The first alternate specification accounts for the argument that impacts may be attributable to a greater extent to an existing disamenity source rather than a future disamenity source, even if the future disamenity is in closer proximity. Accordingly, this specification imposes a ranking system comprising the postconstruction interaction terms for the distance bands up to 5 km (i.e., PC × 0–1 km, PC × 1–2 km, …, PC × 4–5 km), followed by the announcement interaction terms up to 5 km. Under this specification, for the example given above for a property in close proximity to three wind farms, the expected impact of the wind farm that is 2 to 3 km away in the postconstruction period would be more prominent than that of the unconstructed turbine 1 to 2 km away for the wind farm in the announcement period. As a result, in this case only PC × 2–3 km would be categorized as 1, while AN × 1–2 km and PC × 4–5 km would be categorized as 0.
The second alternate specification relaxes the assumption that an impact is attributable primarily to only one (i.e., the closest) wind farm and the corollary requirement that only one interaction term is equal to 1, and allows for impacts associated with the nearest wind farm in each of the announcement and postconstruction periods, whereby interaction terms for both periods can be equal to 1. This allows for accounting separately for announcement impacts and postconstruction impacts, which may differ. Under this specification, for the example given above, both PC × 2–3 km and AN × 1–2 km would be categorized as 1.
Turbine Impacts in Opposed and Unopposed Municipalities
To account for differences in impacts between municipalities that have declared their jurisdictions to be “unwilling hosts” for wind energy development and those that have not, the variables accounting for the turbine impacts are interacted with categorical variables representing the two groups of properties: those located in unwilling host municipalities, referred to in this paper as “opposed municipalities,” and those located in municipalities that have not passed resolutions expressing their opposition to the siting of wind turbines within their jurisdiction, referred to as “unopposed municipalities.”6 This breakdown is based on the list of unwilling hosts compiled by Wind Concerns Ontario.7 As evident in Figure 2, a large proportion of municipalities in the study area of southwestern Ontario (the area west of Toronto and bounded by the Great Lakes) are unwilling hosts; of the 59 municipalities included in the data, 44 are unwilling hosts. As a result, the opposed subsample (comprising 19,683 sales) is much larger than that of the unopposed subsample (2,476 sales). Accordingly, there are considerably fewer sales in close proximity to turbines in the unopposed subsample. For example, of the 300 sales in the 0 to 1 km distance band in the postconstruction period, 38 are located in unopposed municipalities. The number of sales within each distance band is broken down between the two subsamples in Table 1.
Impacts of Turbine Density
Property values may be impacted not only by proximity to the nearest turbine but also by the number of turbines in close proximity. In fact, the disamenity impact of wind turbines may increase as the density of turbines in close proximity increases. To account for the variation in impact with the number of nearby turbines, variables are specified that represent turbine density (number of turbines) within specific distances, namely, 1 km, 2 km, and 5 km. Since the potential disamenity associated with turbine density would be most prominent in the postconstruction period (when the visual impact of multiple turbines on the landscape can actually be observed), these density variables are interacted with the postconstruction time period variable. Specifying densities within different distance ranges allows for examining how the impact of turbine density varies with proximity to these turbines. It is anticipated that the impact of a specific density would be greater for smaller distance ranges, since the turbines would be in closer proximity. The number of turbines within each of the distance limits for properties sold in the postconstruction period ranges from 0 to 7 within 1 km, from 0 to 24 within 2 km, and from 0 to 113 within 5 km. Hence, given the magnitude of turbine density in close proximity to many of the properties in the data set, it is important to assess the impact of turbine density on property values. This involves estimating marginal effects of additional turbines for each of the three distance ranges. The impacts of turbine density are then estimated separately for opposed municipalities and for unopposed municipalities, using interaction terms between the density variables and categorical variables representing each of the two groups of municipalities. The estimation of the impacts associated with turbine density provides a valuable complement to the analysis based on the impacts associated with proximity to the nearest turbine.
Control Variables
Control variables are included in the model to account for the value associated with specific attributes of the property, particularly those related to the house. Attributes accounting for the value associated with the house include square footage, basement area, age of the house, quality level (on a scale of 1 to 10, as rated by MPAC), the numbers of stories, bathrooms, bedrooms, and fireplaces, and the existence of a pool and of air conditioning. Other attributes of the property include the lot size, existence of a garage, and the area of other structures on the property, such as garden sheds. The relative value of specific property types is also accounted for by including variables to represent mobile homes, seasonal properties, and waterfront properties, which are broken down into two categories: lakefront properties and other waterfront properties. Location variables are included to account for urban influence and for the impacts of surrounding property types. Urban influence is accounted for by the distance to the Greater Toronto Area, to the nearest city, and to the nearest major highway interchange. Variables are also included to account for abutting commercial and industrial properties. Summary statistics for the control variables are provided in Table 2 for the full data set, and are also broken down by type of municipality (i.e., opposed and unopposed).
Summary Statistics for the Dependent and Control Variables for the Full Data Set (n = 22,159) and for Opposed Municipalities (n = 19,683) and Unopposed Municipalities (n = 2,476)
Spatial and temporal fixed effects variables are also included in the model. The temporal fixed effects include sets of year and month categorical variables to account for annual and seasonal variation in property values. Spatial fixed effects, specified at the township level, are included to account for unobserved local factors that can influence property values, which can bias the results of hedonic models. The use of spatial fixed effects can substantially reduce omitted variable bias (Kuminoff, Parmeter, and Pope 2010). An alternative approach to addressing this issue is through the use of spatial econometric methods, which account for the spatial dependence that contributes to omitted variable bias. As a robustness check, a spatial lag model is estimated to examine whether potential omitted variable bias may be influencing the results.
3. Methods
Following a number of previous related studies (Vyn and McCullough 2014; Hoen et al. 2015; Heintzelman, Vyn, and Guth 2017), a hedonic analysis is used to estimate the impact of wind turbines on property values, based on a difference-in-differences approach. The empirical model is represented by the following equation:
[1]
where Pijt is the sale price for property i in township j and in time t; Ti is a vector of categorical variables that indicate the time period for the nearest wind farm at the time of sale (i.e., preannouncement, announcement, and postconstruction); Di is a vector of distance band categorical variables (i.e., 0–1 km, 1–2 km, 2–3 km, 3–4 km, 4–5 km, and 5–20 km) that indicate the location of the property with respect to the distance from the nearest turbine; Xi is a vector of control variables, as discussed in the previous section and summarized in Table 28; τt and ρj represent the temporal and spatial fixed effects, respectively; and μjt and εijt represent the local and individual error terms, respectively.
Since the study area encompasses multiple wind farms, the hedonic model represented by equation [1] is used to estimate an average impact on surrounding property values across these wind farms. Pooling all transactions across this area rather than estimating impacts separately for each wind farm allows for basing this estimated impact on a large number of transactions, which may enhance the ability to detect significant impacts, particularly since the impacts of wind turbines tend to be relatively small in magnitude. Conversely, estimating separate impacts for each wind farm may be less likely to generate significant impacts or an accurate estimate of these impacts, as there are relatively few sales in close proximity to many of the individual wind farms.
While combining observations across multiple wind farms may preclude the ability to account for varying impacts across wind farms, this may be less of an issue given the geographic proximity of the wind farms (i.e., all are within southwestern Ontario). In addition, due in part to the short period of time during which most of the wind farms in Ontario were constructed, the turbine sizes are relatively uniform across the province. This contrasts the settings for other recent studies such as those by Gibbons (2015) and Dröes and Koster (2016), where turbine height and capacity vary considerably. As a result of this homogeneity, the impact on property values in southwestern Ontario is unlikely to vary across wind farms due to differences in turbine characteristics.
However, there may be variation across the study area in the values associated with each of the property and location control variables, which may influence the estimated turbine impacts. To account for this variation, the property and location variables (Xi) are interacted with county categorical variables (Ci). Further, in an effort to ensure that the treatment and control groups comprise properties within similar market areas, sales of properties greater than 20 km from the nearest turbine are excluded from the analysis.
The analysis for this study comprises three components. In the first component, equation [1] is used to estimate the average impacts of turbines at various distances on property values across southwestern Ontario. For the second component of the analysis, the variables accounting for the impacts of turbines (Ti·Di) are interacted with each of the two groups of properties: those located in opposed (i.e., unwilling host) municipalities (M1) and those located in unopposed municipalities (M2). The model in equation [1] is then estimated with the two sets of time period–distance band interaction terms (Ti·Di·Mi). Postestimation tests are conducted to determine whether the differences in the estimated impacts between the two groups of municipalities are statistically significant. Finally, given the large numbers of turbines that exist in certain areas of the province, the third component estimates the impact of turbine density on property values, as an alternative to the impact of proximity to the nearest turbine. This involves replacing the interaction terms (Ti·Di) in equation [1] with a variable that accounts for the (logged) number of wind turbines in close proximity. For comparison purposes, the impacts of turbine density and the marginal effects of additional turbines are estimated based on the numbers of turbines within 1 km, 2 km, and 5 km. The turbine density variables are then interacted with each of the two groups of municipalities to examine for differences in the impacts of turbine density between opposed and unopposed municipalities.
The difference-in-differences specification is conducted following the approach used by Hoen et al. (2015), which is based on the model results for the set of interaction terms of time period and distance band variables (i.e., Ti·Di). The interaction term for the preannouncement period and the 5 to 20 km distance band is omitted from the model and represents the control group, as impacts are not expected to occur within this distance range or during this time period. For each of the other distance bands, the impacts in the announcement and postconstruction periods are estimated based on the change in property values that occurred within the band since the preannouncement period relative to the change in property values that occurred over the same period in the 5 to 20 km band. For example, the impact in the 0 to 1 km band in the postconstruction period is calculated as the difference between the change in property values that occurred between the preannouncement and postconstruction periods in the 0 to 1 km band and the change in property values over the same period in the 5 to 20 km band. From the model results, this would be calculated by taking the coefficient for PC × 0–1 km and subtracting the coefficients for PA × 0–1 km and PC × 5–20 km. Similarly, the impact in the 0 to 1 km band in the announcement period is calculated from the model results by subtracting the coefficients for PA × 0–1 km and AN × 5–20 km from the coefficient for AN × 0–1 km.
4. Results
Turbine Impacts
The model results for the interaction terms accounting for the impacts of turbines on property values are provided in Table 3. Based on these coefficients, the difference-in-differences approach described in the previous section is used to estimate impacts on property values in the announcement and postconstruction periods within each of the distance bands up to 5 km from turbines. It is evident from the final column of Table 3 that negative impacts are found up to 4 km from the nearest wind turbine in both the announcement and postconstruction periods. These impacts range from 4.14% to 7.91% in the announcement period and from 5.33% to 8.35% in the postconstruction period. Hence, these impacts do not vary substantially in magnitude across this area, and a decline in magnitude with distance from turbines is not observed, which suggests that impacts may be similar across the area within which turbines are (or will be) potentially visible. These impacts, particularly at greater distances from the turbines, may also be exacerbated by expectations or perceptions of potential impacts that may be attributable to the widespread controversy and public attention in the province regarding this issue.
Estimated Coefficients for the Interaction Terms (Time period × Distance band) and Resulting Turbine Impacts on Property Values
The results in Table 3 also indicate that the estimated coefficient for the 0 to 1 km distance band in the preannouncement period is negative and statistically significant, which is unexpected. This suggests that in some cases wind farms may have been constructed in areas with relatively lower house prices.9 However, despite the lower house prices in the preannouncement period, the results of the difference-in-differences approach indicate that prices are further reduced in the subsequent announcement and postconstruction periods.
Control Variables
The results of the control variables are provided in Appendix Table A1. Since the control variables are interacted with county categorical variables, rather than including an exhaustive list of the results for each of the 14 counties, this table provides a summary of these results. First, for each variable it indicates the percentage of counties for which coefficients have the expected sign and are statistically significant at the 10% level or lower. Descriptive statistics are then provided for these coefficients, which indicate the magnitude of the mean as well as the range of coefficient values. Variables for which coefficients have the expected sign and are statistically significant in at least 75% of the counties include lot size, age, living area, basement area, number of bathrooms, house quality, and lakefront property. Coefficients for several other variables have the expected sign and are statistically significant in at least 60% of counties.
The results for many of the control variables are consistent across counties in the direction of impact on property values. Exceptions include numbers of bedrooms and fireplaces, mobile home, and the distance variables. While bedrooms were expected to positively impact values, the lack of consistency in the direction, as well as the lack of significance, of the coefficients for bedrooms could be due to the fact that with living area held constant an additional bedroom may result in smaller rooms. As a result, the impact of an additional bedroom on the value of a house is ambiguous. The exception that occurs for the impact of a mobile home is an unexpected positive impact in one county (Lambton); however, the coefficient estimate in this county is based on only one observation. Similarly, the impact of fireplaces is found to be positive in all counties except for one (Simcoe).10 The directions of effects for the remaining variables are generally consistent with expectations.
Robustness Checks
The robustness of these results is examined under two alternate specifications of the interaction terms that account for the turbine impacts, in order to address the potential bias due to the assumptions that are imposed in the original specification of these terms. The results of the interaction terms for both alternate specifications are provided in Appendix Table A2, along with the estimated percentage impacts on property values. In the interest of space, the results of the control variables, which are very similar to those of the original specification, are not provided in this table.11 For the first alternate specification, significant negative impacts are found in the announcement period in the 0 to 1 km band and between 2 and 4 km from the nearest turbine, and in the postconstruction period between 0 and 4 km. Similar results are found under the second alternate specification, with an additional significant impact found in the 4 to 5 km band in the announcement period. The magnitudes of the impacts for the alternate specifications are similar to those observed in the primary model.
Another robustness check involves the use of a spatial model to account for spatial autocorrelation, which may affect the accuracy and validity of the hedonic model results. To determine whether spatial dependence exists, a Moran’s I test is conducted. The results of this test (I = 0.105; p < 0.001) indicate that the null hypothesis of no spatial autocorrelation is rejected. This issue can be addressed through the use of either a spatial lag model or a spatial error model, depending on the nature of the correlation (Kim, Phipps, and Anselin 2003). With a higher log-likelihood value for the spatial lag model (3,749.162) than the spatial error model (3,117.296), the spatial lag model is used to account for spatial autocorrelation.12 The spatial weight matrix for this model is specified based on the 10 nearest neighbors.13 The number of observations in the spatial lag model (18,045) is lower than in the ordinary least squares model, as multiple sales of the same properties are omitted to avoid including observations with the same coordinates in the specification of the spatial weight matrix. In cases of multiple sales, only the most recent sale remains in the data, following the approach used by Hoen et al. (2015).
The results of the spatial lag model, provided in Appendix Table A2, indicate that the nature of the estimated impacts is similar to that of the primary model, where significant negative impacts are found within 4 km of turbines in both the announcement and postconstruction periods. This implies that spatial autocorrelation has not significantly influenced the results with respect to the turbine impacts. Similarly, Hoen et al. (2015) found that although spatial dependence existed it did not bias the results for the variables of interest, as the estimated coefficients in the spatial models were quite similar to those of the ordinary least squares models.
Hence, the results of the primary model appear to be robust to alternate specifications of the interaction terms that account for the turbine impacts as well as to the use of a spatial model to account for potential bias associated with spatial dependence. This robustness enhances confidence in the results. The similarity in results between the original specification of the interaction terms and the alternate specifications could be due in part to the relatively low number of properties that are within 5 km of multiple wind farms, particularly where sales occur in different time periods. Of the 4,619 sales of properties within 5 km of at least one wind farm, only 346 sales occurred during both the postconstruction and announcement periods for wind farms within 5 km. As such, the interaction terms for the majority of observations in close proximity to turbines are not affected by the imposed ranking system.
Opposed Municipalities versus Unopposed Municipalities
To examine for differences in impacts associated with differences in attitudes toward wind energy, as identified at the municipality level through declarations as unwilling hosts, the original model is updated by interacting the set of turbine impact variables with categorical variables representing properties in opposed municipalities and properties in unopposed municipalities. Following the estimation of this model, Wald tests are conducted to examine for statistically significant differences in the estimated impacts within each distance band and time period interaction. It is evident from the results of the interaction terms and the Wald tests, provided in Table 4, that differences exist in the estimated turbine impacts between the two types of municipalities. Significant negative impacts of turbines on property values are found in opposed municipalities between 0 and 4 km from the turbines in both the announcement and postconstruction periods, while no significant impacts are found in unopposed municipalities. The magnitudes of the impacts in opposed municipalities are greater than those in the original model, ranging from 5.61% to 9.10% in the announcement period and from 7.93% to 9.42% in the postconstruction period.
Results of the Interaction Terms Accounting for Turbine Impacts on Property Values for Properties in Opposed Municipalities and for Properties in Unopposed Municipalities, with Wald Tests for Differences in Impacts
The results of the Wald tests indicate that a number of statistically significant differences are found in the estimated impacts between the two groups of municipalities. Specifically, significant differences in impacts are found in the announcement period for the 1 to 2 km, 2 to 3 km, and 3 to 4 km (at the 10% level) distance bands and in the postconstruction period for the 0 to 1 km, 1 to 2 km, and 2 to 3 km (at the 10% level) distance bands. Hence, in most cases in which significant impacts are found in opposed municipalities, these impacts are significantly greater than those in unopposed municipalities.
Overall, the results of this component of the analysis provide evidence to support the hypothesis that the nature of impacts of wind turbines on property values is influenced by attitudes toward wind energy, as turbines are found to impact property values in municipalities that indicated opposition to wind energy development but no significant impacts are found in municipalities that have not indicated opposition. Further, the differences between the two groups of municipalities in these estimated impacts are statistically significant in a number of cases.
However, it should be noted that there are relatively few sales within each distance band in the subsample of sales in unopposed municipalities (see Table 1), which may reduce the ability to detect significant impacts. As evident in Table 4, the standard errors for the interaction terms in the unopposed model are about twice the size of those in the opposed model. Despite this potential limitation, the fact that several of the estimated impacts in opposed municipalities are found to be significantly greater than those in unopposed municipalities supports the implication of the model results that impacts are greater in opposed municipalities than in unopposed municipalities.
Turbine Density Impacts
The final component of the analysis examines the impacts of turbine density on property values and estimates marginal effects14 of additional turbines within specific distance ranges. The results of this analysis, provided in Table 5, indicate that turbine density within all three distance ranges (1 km, 2 km, and 5 km) significantly impacts property values, such that higher densities contribute to greater negative impacts. This is consistent with expectations, as multiple turbines within a property’s viewshed would increase the visual disamenity. As evident in Table 5, the marginal effect is greatest for the first turbine and declines considerably for each additional turbine within the specified distance range, which is also consistent with expectations. For example, the marginal effect of the first turbine within 5 km is $4,069, while the marginal effect of the fifth turbine is $1,031 and the marginal effect of the fiftieth turbine is only $106. The marginal effects are also greater in magnitude for smaller distance ranges, as the marginal effect of the first turbine within 1 km is $8,141, or 3.5% of the average sale price of $230,725, which is approximately double the marginal effect of the first turbine within 5 km, while the marginal effect of the first turbine within 2 km is $6,382.
Results of the Turbine Density Variables for Specific Distance Ranges and Marginal Effects of Additional Turbines
Cumulative impacts on property values can also be determined for a specific number of turbines within each distance range. For example, a turbine density of 10 within 1 km would reduce property value by $26,851, or 11.6% of the average value, a density of 10 within 2 km would reduce property value by $21,271, or 9.2%, and a density of 10 within 5 km would reduce property value by $13,746, or 6.0%.
Turbine density impacts are also estimated separately for opposed municipalities and unopposed municipalities to determine whether these impacts differ between the two groups of municipalities. Similar to the results of the analysis based on distance to the nearest turbine, turbine density is found to significantly impact property values in opposed municipalities (see Table 5), while in unopposed municipalities significant impacts are found only for the 5 km density. Once again, the magnitudes of the impacts, as well as the marginal effects, are greater in opposed municipalities than across all sales. The marginal effect of the first turbine in opposed municipalities is $9,342, or 4.1%, for the 1 km distance range, $7,037, or 3.1%, for the 2 km range, and $4,333, or 1.9%, for the 5 km range. The results of postestimation tests indicate that the impacts in opposed municipalities for all three densities are significantly greater (at the 10% level or better) than those in unopposed municipalities. Hence, these results support those of the analysis based on distance to the nearest turbine, which enhances confidence in these results. These results also provide further evidence of differences in turbine impacts across jurisdictions with different attitudes toward wind energy.
5. Conclusions
The results of this study provide strong evidence that wind turbines in Ontario have negatively impacted surrounding property values. The results also demonstrate that these impacts increase with the number of turbines in close proximity. Hence, this study adds to the evidence contributed by more recent empirical studies that wind facilities can impact property values. This study then takes the analysis a step further by examining for differences in impacts among jurisdictions with different attitudes toward wind energy. The observed differences in estimated impacts on property values between “unwilling host” municipalities and unopposed municipalities suggest that the nature of turbine impacts within a jurisdiction (i.e., negative impacts or lack of impacts) may be influenced by attitudes toward wind energy in that jurisdiction, and that public perception regarding wind energy and its potential impacts is a prominent contributing factor to the nature of observed impacts on property values. This is the first study to demonstrate such differences and to provide empirical evidence that may help to explain the lack of consensus in the literature regarding impacts of wind turbines on property values.
With a large number of sales in close proximity to turbines, this study overcomes one of the primary concerns associated with earlier studies on this issue. However, this issue may yet arise in the component of the analysis that breaks down impacts between opposed and unopposed municipalities, as the numbers of sales within specific distance bands for unopposed municipalities are somewhat low, ranging from 29 to 65 for distance bands within 5 km of turbines in the announcement and postconstruction periods. This may represent a potential limitation of this analysis, as the lack of significant impacts observed for properties in these municipalities may be attributable in part to this issue. However, these results are supported by those of the turbine density analysis for unopposed municipalities, which is not affected by the low numbers issue since the postconstruction observations are not divided into distance bands.
The assumptions imposed in the specification of variables accounting for turbine impacts also has the potential to influence the results, which may be a caveat of this study. But once again, the general result of this specification (i.e., turbines negatively impact values) is supported by the results of the turbine density analysis, for which such potentially limiting assumptions are not imposed. In addition, the results under alternate specifications of these variables are quite similar. It should also be noted that these variables are specified based on distance to the nearest turbine as a proxy for the visual disamenity; however, views of turbines may be affected by landscape features such as trees and buildings. While other recent studies have used digital elevation tools to estimate turbine visibility (e.g., Gibbons 2015), which can enhance the disamenity measure based on distance, there are limitations associated with the available elevation data for the study area that inhibit the ability to accurately measure the visibility of turbines from individual properties. As a result, turbine impacts are estimated based solely on a distance measure. The existence of better quality view data for the study area could potentially allow for improving the precision of these estimated impacts.
Another caveat of this study is that the declaration by municipalities to be unwilling hosts is assumed to be representative of attitudes toward wind energy (i.e., opposition to wind energy) for residents of the municipality. However, it is unlikely that all residents within unwilling host municipalities are opposed to wind energy, or that all buyers of residential properties adjust their bids to account for the perceived disamenity associated with turbines. Nonetheless, it is most likely reflective of the predominant sentiment among residents of the municipality, given that these declarations were made in response to public input and feedback from residents regarding this issue. Conversely, such declarations may not have been made by unopposed municipalities due to a lack of public opposition to wind energy within these municipalities and the subsequent lack of pressure from residents to have their municipalities declared to be unwilling hosts.
The finding of significant negative impacts of wind turbines on property values in Ontario is consistent with the results of recent studies such as those by Gibbons (2015) and Dröes and Koster (2016), but differs from the results of Hoen et al. (2015) and Lang, Opaluch, and Sfinarolakis (2014), which did not find significant impacts. The significant negative impacts found in this study extend approximately 4 km from turbines in both the announcement and postconstruction periods. This range of impacts is similar to that of Dröes and Koster (2016), where negative impacts on property values were found to extend 2 to 3 km from the turbines. The magnitude of the impact on property values in Ontario is found to range from 4.14% to 8.35% within 4 km of turbines. These impacts are similar in magnitude to those estimated by Gibbons (2015), who found an impact of 5% to 6% on properties within 2 km of wind farms. Contrary to the results of Gibbons (2015), the magnitude of impacts estimated in this study is not found to decline with distance from the turbines over the 4 km extent of significant impacts, although a decline does occur beyond 4 km.
The results of this study contrast those of a previous study on property value impacts in Ontario by Vyn and McCullough (2014), the results of which indicated no significant impacts of turbines. However, this difference in results is not entirely surprising, due to differences in the settings and time periods between these two studies. First, the study by Vyn and McCullough (2014) examined only one wind farm, which was the first industrial wind farm constructed in the province. The host municipality chose to have this wind farm constructed, and at the time there were relatively few concerns expressed locally regarding potential impacts of wind turbines. The lack of significant impacts on property values surrounding this wind farm is consistent with the results of the current study for unopposed municipalities. Second, the time periods examined by the two studies differ, as the current study includes sales up to 2013, while the study period of Vyn and McCullough (2014) extends only to 2010. As such, the current study includes more of the time period during which public attention and controversy regarding this issue increased substantially in the province. Given the increasing opposition that has occurred over time, it may be the case that general attitudes toward wind energy as well as perceptions of the potential impacts on property values have changed in the province since the construction of the wind farm studied by Vyn and McCullough (2014), which may also contribute to the difference in results.
These differences between the two studies provide further support to the argument that the setting plays an important role in the nature of observed property value impacts of wind turbines. In addition, it is evident from the results of the analysis comparing impacts in opposed municipalities and unopposed municipalities that the nature of turbine impacts varies across jurisdictions, even within the same province. Hence, for jurisdictions considering wind energy development that are concerned about property value impacts, an assessment of prevailing attitudes toward wind energy within the jurisdiction may provide an indication of whether property valueswill be significantly impacted by future wind facilities.
Acknowledgments
This paper is based, in part, on data provided by the Municipal Property Assessment Corporation. Any opinions, findings, conclusions, or recommendations expressed in this material are those solely of the author and are not necessarily the views of the Municipal Property Assessment Corporation.
Footnotes
↵1 It should be noted that these differences in attitudes could occur due to differences in the stage of wind energy development (i.e., proposed vs. existing). Similarly, there could be differences between general attitudes toward wind energy and attitudes toward a specific project. While these distinctions are important to make for understanding factors that influence attitudes and that contribute to differences in these attitudes, the key result from the perspective of the current study is that differences in attitudes have been found to exist across jurisdictions.
↵2 There may also be other factors that influence perceptions of wind energy, such as demographic or socioeconomic factors. Developing a better understanding of the primary factors that influence these perceptions would be a useful topic for future study.
↵3 See https://www.mpac.ca.
↵4 This did not affect the nature of the results with respect to the estimated turbine impacts, but it did improve the model fit.
↵5 Any impact of turbines from other nearby wind farms would be more likely captured through the impact of turbine density, which is addressed later in this paper.
↵6 This designation does not imply that all residents of these municipalities are unopposed to wind turbines, but rather that the municipality has not passed resolutions declaring their opposition. This suggests that there is not a critical mass of residents opposed to wind turbines within any of these municipalities.
↵8 Categorical variables and continuous variables with relatively low integer values are not logged.
↵9 To determine whether factors such as demographic characteristics contributed to this effect, an additional model was tested that included average household income and population density. However, these factors were not found to influence house prices, and their inclusion in the model did not change the results.
↵10 It should be noted that there are only 121 observations in Simcoe County (0.5% of all observations).
↵11 However, these results are available from the author upon request. This is also the case for all subsequent models estimated in this paper, for which only the results of the interaction terms are provided.
↵12 However, the results are similar for both the spatial lag and spatial error models.
↵13 An alternate spatial weight matrix, based on an inverse distance specification, is also tested. The results are consistent across both specifications.
↵14 Marginal effects are estimated using the “margins” postestimation command in Stata.
