THE EFFECT OF LAKE WATER QUALITY AND WIND TURBINES ON RHODE ISLAND PROPERTY SALES PRICE

This dissertation uses the hedonic pricing model to study the impact of lake water quality and wind turbines on Rhode Island house sales prices. The first two manuscripts are on lake water quality and use RI house sales transactions from 1988-2012. The third studies wind turbines using RI house sales transactions from 2000-2013. The first study shows that good lake water quality increases lakefront property price premium. It also shows that environmental amenities, such as forests, substitute for lake amenity as the property’s distance from the lake increases. The second lake water quality study incorporates time variables to examine how environmental amenity values change over time. The results show that property price premium associated with good lake water quality does not change as it is constant in proportion to housing prices with short term economic fluctuations. The third study shows that wind turbines have a negative and significant impact on housing prices. However, this is highly location specific and varies with neighborhood demographics. All three studies have policy implications which are discussed in detail in the manuscripts below.


LIST OF TABLES
MANUSCRIPT 1: The Impact of Lake Water Quality on Rhode Island Property Values

LIST OF FIGURES
MANUSCRIPT 1: The Impact of Lake Water Quality on Rhode Island Property Values lakefront property sales prices based upon level of lake water quality are estimated using a single lake as well as all Rhode Island lakes scenarios. The improvement only from marginally poor to good water quality shows significant potential benefits. The findings from this study on homeowner preference for lake water quality provide important information for RI policymakers.

INTRODUCTION
Lake water quality impacts both consumptive and non-consumptive uses by the surrounding community (Wilson and Carpenter, 1999;Millennium Ecosystem Assessment Report, who.int 2005). Federal and state governments have allocated substantial annual investments to preserve and improve its integrity through surface water management (U.S. EPA). Despite that, sixty six percent of lakes and reservoirs in the United States were classified as impaired for one or more of their designated uses in 2009 (Walsh et. 2011). Since Rhode Island has over two hundred sizeable lakes 1 , maintenance of lake water quality is important to Rhode Island communities.
This study uses chlorophyll concentration as a water quality indicator to examine the impact of lake water quality on surrounding property sales prices. Close proximity to an environmental amenity is generally incorporated as a property sales price premium from which amenity value of lake proximity can be quantified. As part of the assessment of natural amenities' effect on property sales prices, the study selects forests in the vicinity of lakes as an additional environmental amenity to examine the significance to nonconsumptive amenity value by nearby property owners.

DEVELOPMENT OF HEDONIC MODELS
Hedonic pricing models have been used for characterizing the prices of competitively traded goods comprised of heterogeneous sets of characteristics. Housing market is most commonly used for environmental hedonic models because of common spatial factors 1 These lakes are larger than 10 acres in size and tracked by RI DEM.
such as location of houses and their surrounding environmental attributes. A given housing unit is best characterized as consisting of a bundle of attributes that in aggregate describe the structure itself, the land upon which it is built, and the relevant location characteristics. The hedonic approach attempts to separate the internal property attributes (baths, bedrooms, square feet, etc.) from the public and private good attributes associated with location. For example, hedonic pricing model identifies a premium paid for houses located near desirable environmental amenities, according to the premise that price is determined by both internal characteristics of the good being sold and external factors such as many environmental externalities (Freeman, 2003). In this study, the environmental amenity (lake water quality) is a characteristic, a non-market good, and the market good is a house.

PREVIOUS STUDIES
The existing empirical research with hedonic price model to determine the value of water quality is fairly limited in contrast to the abundant literature on other environmental goods such as air quality. Although degradation of either water or air quality may adversely impact nearby property values (Walsh et al. 2011), the limited literature on water quality is may be due to the lack of consistent and accurate water quality data available to homeowners (Kashian et al. 2010). In addition, the latent and idiosyncratic nature of water quality poses a challenge to find an appropriate water quality indicator (Legget et al. 2000).
Secchi Disk Measurement (SDM) is a relatively easy measurement method which is based on visibility, and is a frequently used water quality indicator (Epp and Al-Ani, 1979;Boyle et al, 1999;Michael, Boyle and Bouchard, 2000). It has been used alone or in combination with other indicators (Poor et al. 2001;Boyle et al. 1999). Multiple water quality indicators are explored as well. For example, Walsh et al (2009)  Properties in close proximity to environmental amenities are a consistent focus for hedonic environmental studies; lake water quality study is no exception. One of early studies uses subjective qualitative water quality ratings (poor, moderate, and good) to examine the correlation between lakefront property prices and water quality of artificial lakes in Wisconsin (David, 1968). Her findings show a correlation between subjective water quality ratings and lakefront property prices, but studies utilizing objective measures of water quality are more useful (e.g., Leggett and Bockstael, 2000;Kashian et al. 2010). Changes in objective measures can potentially be forecast to estimate benefits associated with policy changes. It may not be possible to do the same for subjective measures.
A study using 39 objective measures in Lake Michigan finds that lakefront property prices are capitalized into property sales prices by observable water quality measures (Brashares, 1985). Other studies with an assortment of multiple water quality indicators on riverfront properties report a similar relationship (Epp and Al-Ani, 1979;Legget and Bockstael, 2000). Young and Teti (1984) examined the degraded water quality in St.
Albans Bay, Vermont to find that properties closest to the degraded Bay suffered the most and are consistent with the findings from studies on water quality improvement (Michael, Bouchard, 2000, andPoor et al., 2001). Some studies combined Hedonic model with a survey based approach. For example, Boyle et al (1999) combined hedonic method and survey of 500 homeowners in proximity to thirty three lakes in Maine and found the similar correlation between water quality and property pricing.
The distance effect of water quality on property price is another common focus of lake amenity studies. One study estimates the marginal amenity value from the difference between within and outside of a 2000 ft. distance threshold (Lansford and Jones, 1995).
Their findings demonstrate that the marginal value trend for the lake water level per foot diminishes with distance from the lake to non-significance beyond the threshold. A comparison study shows that water quality amenity value benefits non-lakefront as well as lakefront properties (Walsh et al., 2011). Another distance differentiation study suggests that other landscape attributes become important and may replace lake amenity beyond its distance threshold (Palmquist and Fulcher, 2006).
Some studies have found that lake water quality is highly susceptible to the types of landscape attributes in the vicinity of the lake. For instance, lake water quality correlates with the number of homes surrounding the lake and homeowners' land use practices (Leggett et al., 2000). High housing density is a common source of excess nutrients that exacerbate lake eutrophication. The size of agricultural land and distance from lakes have a significant impact on lake water quality, and this reflects on property values (Bolitzer et al., 2000). An ambient water quality study shows residential development near lakes has a significant effect on lake water quality (Epp et al., 1979). They also find that there may be a threshold effect, such that there is little or no benefit to marginal improvements in water quality for houses in proximity to water bodies of very poor quality water, whereas housing prices are sensitive to water quality improvement for those adjacent to higher quality water. The latter study's findings relating to both consumptive and non-consumptive water use can help lake management policymakers prioritize the lakes that need improvement.
The direct correlation between lake use and lake water quality has elicited many studies to develop management policy. The study by David (1968) on lakefront property values that became a guide for public lake management is followed by many others. Studies have found that homeowners' preference for lakes vary with the types of recreational services available (Boxall et al., 2003;Kaplan,1985;Whitehead et al.,1991;and Poor et al.,2001). Larger lakes can accommodate recreation activities such as boating, canoeing, swimming, fishing, and trails whereas small lakes provide mainly aesthetic benefits (Young et al., 1984). Lakes with boating as the main recreation service may require different water quality criterion compared to lakes with primarily swimming.
Homeowners' preference for recreation use may have significant management policy implication.
Designated uses are the core classification criteria defined by the EPA and regulated against lake water quality. While criterion level of each water quality is objectively measured, subjective measures based on the individual designated users' perception and preference are reflected in their property values.
The hedonic method is based on the assumption that consumers have complete knowledge of or information on the goods they are purchasing and incorporate this information into their buying decision (Freeman, 2003). Yet, individuals' preference and perception vary vastly based on population demographics and affects their buying decisions. While swimmers would prefer a lake with high transparency, recreational anglers value a lake of higher trophic level for better fish habitat (Hoyer et al. 2004).
One study showcases the importance of perception with a lake discolored dark brown with tannic acid. Property values were negatively affected by this subjective aesthetic quality (Steinnes, 1992).
Discrepancy between perceived and actual water quality has been a persistent topic with hedonic lake studies. Consider the study by Poor et al. (2001)

Limitations and Challenges of Hedonic Price Model (HPM)
As is true of all statistically-based analyses with non-experimental data, the hedonic approach faces potential challenges associated with omitted variables, endogeneity, and spatial dependence or autocorrelation.

APPROACH OF THIS STUDY
This study tests the following hypotheses: (1) The amenity value associated with lake water quality differs for lakefront properties versus non-lakefront properties, ceteris paribus.
(2) The amenity value of lake size is affected by lake water quality, ceteris paribus.
(3) Other environmental amenity values become more prominent for properties that are more distant from the lake, ceteris paribus.
This study emulates the distance differential study by Lansford and Jones (1995) with modification of replacing lake water levels that were used in their study to determine amenity value with lake water quality of this study. This study also examines the inference by Palmquist et al. (2006) that as lake amenity values diminish with increased distance from the lake, other environmental attributes become important. The additional environmental amenity used in this study is forests. This study is analogous to Walsh et al. (2011) as it analyzes a large number of property sales transactions and man-made lakes. Both studies also use logarithmic functional forms and interactions terms in hedonic models to explore the water quality effect differentiation between lakefront and non-lake front properties. The high population density and large number of property sales transactions in Rhode Island makes the hedonic model an ideal tool for this study.

STUDY AREA: Rhode Island Single Family Homes near RI Lakes
This study quantifies how lake water quality affects property sales prices for single  This study combines Rhode Island property sales transactions 3 and chlorophyll concentration data 4 with geographic information from Rhode Island Geographic

DATA OVERVIEW
Information System (RIGIS) to estimate the effect of lake water quality on housing prices. Both Rhode Island property sales transactions and lake water quality data are from the same 1988-2012 database. Chlorophyll concentration is chosen for this study because it provides a good overall measure of lake water quality status, especially when excess nutrients are a key water quality issue

GIS dataset
The location of a house is a major determinant of its value (Bourassa, 2006;Theriault et al. 2003). The market value of a house can be expected to reflect nearby environmental amenities, and the effect on housing price is expected to decline with distance to the environmental amenity (David, 1968;Walsh, 2009)

Functional Forms and Model Selections
This study uses the logarithmic functional form. It is shown that econometric models for the equilibrium price function perform best when all variables are included in the model but that simpler functional form using a linear, log-linear specification performed best in the presence of omitted variable (Cropper et al., 1988). Logarithmic and semilogarithmic functional forms which represent the elasticity in percentage render easier interpretations. Linear and squared terms were tested for primary living area, age and lot size because theory and empirical results suggest nonlinearities in valuing these characteristics.
Following tests of alternative specifications, this model is selected: ln (Price ijt ) = λ t + α j + γ k + σ i + β 0 + ∑β 1 X ij + β 2 lakefront ik + β 3 llakeft k + β 4 lforest i + β 5 lacre ik + β 6 lakefront ik *goodWQ ikt + β 7 llakeft k *goodWQ ikt + β 8 llakeft k *lforest i + β 9 lacre ik *goodWQ ikt + ᶓ it , [1] where Price ijt represents the price of the property i adjusted to the second quarter of year 2010 RI housing price index. λ t represents a year-quarter fixed effects; α j denotes the Census tract fixed effects; γ k denotes lake fixed effect; σ i denotes bedroom fixed effects; goodWQ ikt is a dummy variable for the oligotrophic and mesotrophic levels of water quality. X ij represents the house characteristics variables such as living area, number of total rooms, number of bathrooms, exterior condition of the house, number of bedrooms, living area and the age of the house, etc. The variable lakefront is a dummy variable for lakefront properties and defined as properties within 100 meters from the lakeshore 9 .
llakeft is a logarithmic distance in ft. between a property and its nearest lake with chlorophyll measurement. lforest and lacre are logarithmic forest, and lake surface area respectively in square meters within 0.25 miles from a property, and ᶓ it , is error term. Based on the mean house sales price ($263,348), this is equivalent to a premium of $18,487 in combined effect of lakefront property and good lake water quality. The edge effect of lakefront properties declines rapidly beyond a cut-off distance as observed in other studies (Brown and Pollakowski, 1977;Landford and Jones, 1995). The main effect of natural logarithmic proximity to the lake in ft. variable, llakeft, is not significant.

RESULTS
Its interaction term with water quality, llakeft*goodWQ is negative and significant as -0.894% . Its negative net effect indicates that house sales prices decrease as the distance between a property and its nearest lake increases when water quality is good.
TABLE 4 shows that the natural logarithmic forest surface area has a positive and significant impact on house sales price (0.45%) in its interaction term with lake distance, lforest*llakeft. This trend is opposite to that of lake water quality and lake distance interaction term, llakeft*goodWQ, and indicates that forests substitute for lake amenity values as the distance between a property and its nearest lake increases. This observation corroborates the findings from other studies (e.g., Palmquist and Fulcher, 2006).
Lake size is usually included only as an interaction term in regressions because lakes do not change their size. Accordingly, the interpretation of its coefficient as an independent variable warrants a caveat that it may overestimate the significance (Walsh et al., 2011).  4 shows that the lake size variable, the logarithmic surface area of lakes in square meters, lacre, does not have a significant main effect. However, its interaction term with good lake water quality has positive significant effect on house sales price (1.17%). The fact that good lake water quality has a positive effect on the lake size amenity value is consistent with findings from other studies (e.g. Walsh et al. 2011). These lake size observations may be related to the variety of recreational services that larger lakes offer if water quality is adequate to support those activities.

POLICY ANALYSIS
This study's results are applied to hypothetical water quality improvement policies encompassing all RI lakes. Three scenarios are explored: (1) a single lake with poor    6 lists non-good lake water quality variables to estimate the total change in lakefront property sales prices if the lake were to switch from non-good to good water quality. 11 Two non-good water quality levels include poor (eutrophic chlorophyll concentration 7.2 to 35 ppm) and extremely poor (hypereutrophic chlorophyll concentration greater than 35 ppm). The reference water quality level is good water quality, which includes oligotrophic chlorophyll concentration (less than 2.6 ppm) and mesotrophic chlorophyll concentration (2.6 to 7.2 ppm). The lakefront interaction term with poor (eutrophic) quality is significant and is negative, it reduces lakefront property sales prices by 3%. The interaction term with extremely poor (hypereutrophic) water quality is not significant. This is likely due to the small number of observations, with only 230 lakefront property transactions near 5 extremely poor lakes.

CONCLUSION
Good lake water quality is an environmental amenity, as evidenced by its positive impact on neighboring property sales prices. The amenity edge effect on lakefront properties is only for close proximity, likely due to the latent nature of lake water quality. Forest amenities substitute for lake water quality as distance increases, which is consistent with the findings from other studies (e.g. Walsh et al., 2011;Palmquist et al. 2006). Since good lake water quality benefits non-lakefront as well as lakefront properties as shown by edge effect and amenity substitution, the scope of lake water quality management needs to extend beyond the lakefront. Taking into consideration the recreational services that different size lakes may offer, along with homeowners' willingness to pay for recreational services, lake size is another important consideration to incorporate into lake water quality management policy. The estimation of potential increase in lakefront property sales prices shows that the improvement from poor (eutrophic) to good water quality is significant while the improvement from extremely poor (hypereutrophic) to good is not statistically significant. These results are consistent with the findings from other studies (e.g. Epp et. al, 1979)         Year-Quarter, Census Tract and Lake ID Level of Significance codes: 0 '***' 0.001' 0.01 '*'0.05 '.'0.1 ' '1; the number 0.00 denotes numbers less than 10 -6 ; and   Year-Quarter, Census Tract and Lake ID Level of Significance codes: 0 '***' 0.001' 0.01 '*'0.05 '.'0.1 ' '1; and 0.00 denotes numbers less than 10 This study examines the sensitivity of estimated environmental values over time within the context of a case study of the amenity value of water quality in Rhode Island lakes.

FIGURE 4 | A Single Lake for Policy Analysis
The hedonic model is used with data for Rhode Island house sales transactions from 1988 through 2012 to estimate the price premium associated with the lake water quality over this time period. Sensitivity of the price premium over time is estimated using two time-

PREVIOUS STUDIES
Hedonic pricing model quantifies environmental amenity value revealed through the property sales price premium for an environmental amenity in its proximity. This is based on the premise that environmental amenities behave as normal goods. Clean air quality is commonly used in literature to test the hypothesis that environment amenity is a normal good. There is a negative correlation of air pollution intensity by nitrogen and sulfur compounds with income per capita across the U.S regions (Bruneau and Eschevarria, 2003). Looking at another amenity, property owners in proximity to lakes having higher levels of human capital (the proxy being the shares of college graduates) also suggests that environmental amenity is a normal good (Stephens and Patridge, 2012).
Overall, the findings are consistent with the theory that environmental amenities are normal or superior goods.
In general, one would expect income to be an important factor in determining environmental amenity values (Antle et al.,1995;Barbier, 1997;Yandle et al., 2002). An open question is the extent to which time scale matters. Changes in income over the long term can have different effects on environmental amenity values compared with short term fluctuations in income. Friedman's permanent income hypothesis suggests that values are not expected to be very sensitive to short term volatility in income. Taken to an extreme, the permanent income hypothesis could imply that people anticipate long term income increases, and this anticipation may be reflected in environmental amenity values. As a result, environmental amenity values may not be very sensitive even to long term income changes. In contrast, if households behave myopically, one might expect to find the value of environmental amenities to vary with short run changes in income. This provides a rationale for testing the statistical significance of both long term income trends and short term fluctuations in revealed values for environmental amenities.
Real estate plays an integral role in the economy. 2 The real estate recession and boom cycles trend is a good indicator of economic health. It is generally considered that during weak economic conditions consumers are forced to alter their financial decision making with different spending patterns (Shahid, 2008). Since real estate is a large portion of the typical consumer's expenditure, economic conditions affect spending patterns related to both disposable income and larger, long-term financial assets (Stein, 1995). Furthermore, environmental amenities are often considered to be luxury goods. Since change in demand for environmental amenities to be more than proportionate to income changes, the study of how economic cycles affect environmental amenity value warrants an analysis of both demand elasticity for environmental amenities as well as income elasticity (Martinez-Alier, 1994;Bruneau et al, 2003).
The concept that the demand for environmental amenities is sensitive to changes in income underlines the Environmental Kuznets Curve (EKC). Under the EKC, environmental quality decreases with increasing income until a threshold income level is reached, after which environmental quality improves with income level (Barbier, 1997).
This result is based on the notion that if environmental quality is a luxury good, then demand for environmental quality increases as incomes get sufficiently high. (Antle and Heidebrink, 1995).
Numerous studies explore EKC in broad, global perspectives (e.g. Stern et al 1996). A more recent study focusing on a specific environmental amenity assesses the income elasticity demand for environmental quality in Sweden using recreational services as an environmental quality indicator (Ghalwash, 2008). It confirms that recreational services are a luxury good. The study also includes other traditional groups of goods for the analysis of how the income elasticities for these composite goods change over time. recession. They find that the environmental amenity is a normal good, as marginal implicit values decreased during recession and increased during boom. A study examining the relationship between county per capita income and toxic pollutants using a Kuznets Curve model (Rupasingha et al., 2004) corroborates the findings of Cho et al. (2009aCho et al. ( , 2011. Other studies using different environmental attributes also show environmental amenity as a normal good: the value of a greenbelt in Seoul changed with the recession-boom cycle (Lee and Linneman, 1998); undeveloped land in proximity to vacant land has a higher value during boom cycles (Smith et al., 2002); and an analysis of data from the 1970's and 1980's shows that consumer marginal willingness to pay for improved air quality was lower during the 1981-82 recession (Chay and Greenstone, 2004).

HYPOTHESES OF THIS STUDY
(i) Environmental amenity values change over time. This study will determine whether these values increase in conjunction with improved economic conditions over a continuous time horizon.
(ii) Environmental amenity value changes vary depending upon the time scale, whether they are short term fluctuations or long term trends. This study is designed to test whether lake water quality impacts house sales price premium more during upward versus downward economic conditions.
(iii) Environmental amenities are luxury goods. Using edge effect of lake water quality on lakefront properties as the environmental amenity, fluctuations in economic condition is expected to affect the price premium of lake front properties more than non-lakefront properties.
The first two hypotheses mirror Environmental Kuznets Curve that models demand for environmental quality depends upon income. Two explanations to consider: exogenous changes in environmental values and environmental values increase over time with income, with the latter more likely to show a response to economic cycles. In addition, response to economic condition is more about the time scale of sensitivity to income changes. One argument would be that current income matters. A second argument would be like a "permanent income hypothesis", which could be a speculative effect or a smoothing over time that prices don't respond to current income because there is an expectation that prices will rebound in the future. The third hypothesis is based on the premise that environmental amenities behave as normal goods and economic conditions influence consumers differently in accordance with their financial status. If environmental quality is a luxury good, one would expect the price premium to be more sensitive to fluctuations in income. The hypotheses test time sensitivity of environmental amenity values in the context of RI lake water quality in response to short term economic conditions using the Rhode Island Current Condition Index (RICCI), and to long term time trend using a liner number of days.

OVERVIEW OF STUDY AREA and DATA
Rhode Island's landscape encompasses more than 5,000 lakes covering 20,749 acres. single family homes within 0.5 miles from their nearest lakes are extracted and they include 3,315 single family lakefront properties. The cut-off proximity for lakefront property is 100 meters from its nearest study lake.

Chlorophyll concentration dataset
Since 1988, the University of Rhode Island has coordinated a volunteer-based lake monitoring program through the URI Watershed Watch (URIWW) program. 14 This program is the primary source of ambient water quality data on lakes in RI. Watershed Watch trains volunteers to collect samples seasonally from May through October at a total of 99 water bodies in Rhode Island. Sample analysis is performed in URI laboratories. Water quality parameters measured in the URIWW lake monitoring program include: water clarity (secchi depth), water depth, temperature, dissolved oxygen (deep sites), pH, alkalinity, chlorophyll, total and dissolved phosphorus, total nitrate, ammonium -nitrogen, chloride and pathogens.
Chlorophyll concentration is chosen for this study because of its visual manifestation of lake water quality status. It also serves as a good summary measure of water quality in Rhode Island lakes that simplifies the complex idiosyncratic nature of lake chemistry and serves as a trophic-level proxy. Chlorophyll concentration is a key water quality measure that reflects eutrophication levels of lakes. Overall, chlorophyll concentration is a good holistic measure, particularly when excess nutrients are a primary water quality issue . Protecting lake water quality by mitigating eutrophication is a major challenge in Rhode Island. 15 Chlorophyll concentration is categorized in four conventionally used trophic levels according to the range of concentration in increasing order: oligotrophic, mesotrophic, eutrophic and hyper-eutrophic. The chlorophyll concentration unit of microgram per liter is equivalent to parts per million, ppm. For this study, a dummy variable, goodWQ is defined as one for oligotrophic and mesotrophic levels, equivalent of chlorophyll concentration less than 7 ppm, and zero otherwise.

Geographic Information System (GIS) dataset
The location of a house is a major determinant of its value (Bourassa, 2006;Theriault et al. 2003). The market value of a house can be expected to reflect nearby environmental amenities. Typically, one would expect that properties in closer proximity to the amenity will have a larger effect, the larger the effect sales price (David, 1968;Walsh, 2011) 16 .
This location dependent externality distribution is important for lake water quality management plans. Accordingly the GIS is instrumental for this study. The unique RI address locator from Rhode Island Geographic Information System (RIGIS) based on Emergency 9-1-1 structure coordinates is used to geocode locations of all the houses with sales transaction and the longitudes and latitudes associated with the housing dataset are corrected accordingly. Their locations are further verified with an intersect tool with RI Census shape files using ArcGIS Intersect (Overlay) Tool.
All datasets are merged using ESRI ArcView GIS software 10.1 and statistical software STATA 12. Both open source software R and statistical software STATA 12 are used for data management and for the hedonic price function estimation that is used to specify the effect of water quality on sales prices.

ENVIRONMENTAL AMENITY AS SUPERIOR GOOD
There are two types of goods in relation to consumer's income: inferior goods and normal goods. FIGURE 4 shows how demand for each type of good behaves with change in income. Demand for inferior goods decreases (from Q to Q 3 ) as consumer's income increases (from I 1 to I 2 ). Thus, the income elasticity is negative. The income elasticity for normal goods is positive since demand increases (from Q to Q 1 ) as consumer's income rises (from I 1 to I 2 ). An extreme form of normal good is superior good. Superior 16 Although this is a general consensus, it might not be true for all environmental amenities. For some amenities such as farmland or forest, we might want to be close, but not too close.
goods make up a larger proportion of consumption (from Q to Q 2 ) as income rises (from I 1 to I 2 ), thus a superior good's income elasticity is both positive and greater than 1. A superior good is said to be a luxury good if it is not purchased at all below a certain level of income. Superior good and luxury good are also normal goods, but a normal good is not necessarily a superior good or a luxury good (Mankiw, 2007).
The bell shape of Environmental Kuznets Curve (EKC) in FIGURE 5 models the hypothesized relationship between environmental quality and income (Barbier, 1997) and it becomes a luxury good at higher levels of income (Antle and Heidebrink, 1995). The demand relationship determines income elasticity, which in turn determines whether a normal good is a necessity or a luxury. Income elasticity may vary with income, but not necessarily. And a good may be a necessity (strictly positive) at all income levels while another good may be a luxury good at all income levels.
Consider a simple example with two goods, X and Y, where X is a superior good and Y is an inferior good as below: x* = X (p x , p y , I ) ; y* = X (p x , p y , I ) , where both x* and y* are the optima demand for respective normal goods, and p is price of good in subscripts. I, the income constraint is defined as I = p x x + p y y. FIGURE 6A demonstrates the changes in income from I 1 to I 2 and vice versa, ceteris paribus prices and preferences (utility function), shift the budget constraint parallel, thus X 1 to X 2 and Y 1 to Y 2 , and vice versa. For normal goods, as income increases from I 1 to I 2 , demand increases from X 1 to X 2 . If income elasticity of demand for X is greater than 1 as shown, then normal good X is a superior good. The decreased demand for good Y from Y 1 to Y 2 as income increases from I 1 to I 2 indicates that good Y is an inferior good. FIGURE 6B is another Engel Curve that shows how demand for normal good y changes as income changes, ceteris paribus, in which y is a function of income, f(I).
Now, suppose these two normal goods are differentiated as luxury and necessity good as shown in FIGURE 7. Q* is optima quantity of good demand and it is a function of income, f (I). For normal goods, or f I Curves may bend up for luxury goods (f II > 0) so that income elasticity is greater than 1, and down for necessities (f II < 0) with income elasticity less than 1. This holds when income is above the threshold so that non-zero quantities of the luxury goods are purchased. Income elasticity is zero for the luxury good when income is strictly below the threshold. Both necessity and luxury goods have a positive income elasticity, these two curves intersect at the threshold income level, I T . Demand for luxury good will decline by more than demand for necessity good below I T . By definition, the derivative of demand for the luxury good with respect to income is greater than derivative for other normal good (necessity good).
Environmental amenity is a luxury good. We would expect that the price premium for high environmental quality will decline in down economic times, and the degree with which economic conditions impact income levels would mirror environmental amenity values.

DEVELOPMENT OF HEDONIC PRICE MODELS
Hedonic pricing models have been used for estimating the values of competitively traded heterogeneous goods. Under appropriate conditions, the hedonic model allows one to identify the contribution that characteristics make to the market price of heterogeneous market goods (Freeman, 2003). Housing market is most commonly used for environmental hedonic models because of common spatial factors such as location of houses and their surrounding environmental attributes. In the context of housing market, characteristics include the structural characteristics of the properties (bathrooms, bedrooms, lot size, etc.) and non-structural characteristics associated with the location, including environmental characteristics. The hedonic model's structural form that internal property characteristics can be decomposed from other non-structural attributes associated with location is the attractive feature for its popular application to study environmental attributes. For example, a hedonic pricing model identifies a premium paid for houses located near desirable environmental amenities, according to the premise that price is determined by both internal characteristics of the good being sold such as structural characteristics of a house, and external factors such as environmental externalities (Freeman, 2003). The environmental amenity of interest in this study (lake water quality) is a non-market characteristic, and the house is the associated market good.

Limitations and Challenges of Hedonic Pricing Model (HPM)
As is true of all statistically-based analyses with non-experimental data, the hedonic approach faces potential challenges associated with omitted variables, endogeneity, and spatial dependence or autocorrelation. In this study, characteristics of the house, property and the neighborhood are the control variables to reduce possibilities of omitted variables. Endogeniety, spatial dependence, and autocorrelation error can be addressed with fixed effects applied at a specified geographic range (Kuminoff, Parmeter, and Pope, 2010). This study uses interactive terms along with local fixed effects to address these issues. Variation between the explanatory variables is measured within the scale of the fixed effects so if we are measuring proximity to various amenities using Census tract, the variation that would otherwise be present between individual parcels is lost. Using fixed effects requires that the variables included in the regression vary over time at the specific level the fixed effects are applied. Chlorophyll concentration measurements over 1988-2011 span provide the requisite variations.

Estimation of the Hedonic Price Function Functional Forms and Model Selections
Since the house sale price is in logarithmic functional form, semi-logarithmic equation is used as functional form in this study. Log-log function form renders easier interpretation of coefficient. Linear and squared terms were tested primarily for living area, age and lot size, but only linear function forms are used for these variables. Numerous trials with functional transformation and model selections render the following equation: ln (Price ijt ) = λ t + α j + γ k + σ i + β 0 + ∑β 1 X ij + β 2 lakefront k *goodWQ ikt +

RESULTS
TABLE 1 shows the coefficient estimates for key variables using house sales transactions for properties within 0.5 mile to the nearest lake, with those within 100 meters identified as lakefront properties. 18 Main effects show lakefront is significant among the variables of interest. The interaction term between lakefront property and good lake water quality, lakefront*goodWQ indicates how lakefront property prices are affected by water quality.
The interaction term between lakefront, water quality and days, lakefront*goodWQ* *days indicates how the water quality effect on property prices changes over time. The coefficient of interaction between lakefront property and good lake water quality, lakefront *goodWQ is positive, but not statistically significant at the + 1.95% level. The coefficient of interaction between lakefront property, good lake water quality and days, lakefront*goodWQ**day200 is also positive and not statistically significant at the 0.0662 % level. The other time variable RICCI exhibits the same trend, but with negative and not significant coefficient. This indicates that the water quality effect does not change over time. These non-significant coefficients indicate that we cannot reject the hypothesis that amenity value for lake water quality is constant in percentage terms over varying economic conditions and over time.
Lake water quality is significant only in small distance between a property and its nearest lake, this includes lakefront properties. Looking at the environmental amenity of good water quality as a superior good, the demand for lakefront properties should increase more than proportionately when income increases. However, the non-significant interaction terms both lakefront *goodWQ** days and lakefront *goodWQ**RICCI suggest that price premium for water quality for lakefront properties 19 is constant in proportion to housing prices. For example, if housing prices increase by 20%, so does amenity value as it stays constant in percentage terms to economic conditions on short time scales. This suggests that people are not myopic on these time scales (months to a couple of years) and this notion warrants a further investigation in future studies.         Because the fear of potential adverse effects of wind turbines on nearby property values has been among the most important concerns (e.g., Bond, 2010), the siting of wind turbines has been an extremely contentious issue. The hedonic method is a promising approach for identifying the extent to which housing prices are affected by nearby wind turbines.

intervals (12years)
Appropriate market definition and scale are prerequisite for effective hedonic analysis; housing market is no exception (Dorsey et al, 2010). Incorporating residential demographics into hedonic model is necessary because of Rhode Island's high housing density and town heterogeneity in a single housing market. Although disproportionate externality for properties in close proximity to wind turbines is not addressed, this study's use of residential demographics helps understand community attitude towards wind turbines.
The potential negative impact on local property owners has been classified into three stigma categories: scenic vista, nuisance, and area (Hoen et al., 2009(Hoen et al., , 2011. Scenic vista stigma results from adverse effect and/or obstruction of views from a property. For example, a wind turbine might adversely affect an otherwise pristine ocean view. Since a scenic view is considered a premium attribute for property value, scenic vista stigma has been commonly examined in hedonic studies of wind energy (e.g., Dent, P. and S. Sims, 2007;Sims et al. 2008;Hoen et al. 2009Hoen et al. , 2011. Nuisance stigma includes noise, infrasonic vibration, and shadow flicker. Nuisance stigma primarily impacts homeowners within a small proximity as they bear disproportioned externality (Pierpont, 2008;Colby et al. 2009;Heintzelman and Tuttle, 2011).
Area stigma includes the perception that the community is an industrial location. This is the most subjective of the three forms of stigma and could be the most far reaching, as this brings the concern from an individual property owner perspective to that of the wider community (Heintzelman and Tuttle, 2011). For this most subjective form of stigma, inclusion of area demographics would enhance understanding of wind turbine effect on community attitude and location impact. This study's combination of residential demographics and RI's unique geographical features helps examine all three stigmas and guide wind energy development.

PREVIOIUS RESEARCH
Parallel to the rapid growth of wind energy development, many studies of wind energy have emerged. Since the potential negative impact by wind energy development on its nearby property values has been a common concern, hedonic method is an appropriate analysis tool to assess the significance of the impact by wind energy development and the role of perception by homeowners at the community level.
The location of and view from a property influence the premium value of a property Nuisance stigma such as shadow flicker, sound, and vibration is pertinent to homeowners in close proximity to wind turbines. Both nuisance and scenic vista stigmas were investigated in the study by Dent, P. and S. Sims (2007) with 919 transactions for homes within five miles of two wind facilities in Cornwell, UK. Despite initial evidence that there was an effect, their further investigation reveals other factors that were more significant than the presence of a wind farm. It was not differentiable from their study whether it is due to the small size of the transactions in close proximity or the impact of the effect. Another study that analyzes scenic vista stigma with 280 residential transactions of homes near a wind facility in Madison County, NY finds no evidence that view of turbines significantly affects home sales prices (Hoen et al., 2006).
The development timeline of wind energy is introduced to examine the anticipatory effect on homeowners' perception of wind energy and post construction effect in the studies (Hoen et al. 2009(Hoen et al. , 2011. These studies render a comprehensive approach by combining spatial and temporal components in the process. Discrete proximity increments and their interaction terms with wind turbine development timeline show that the impact of wind turbines is not statistically significant to nearby property values. They also find that home sales prices within a mile of the turbines more than two years prior to the facilities' announcement and those that sold in later periods (e.g. post announcement or post construction) are statistically indistinguishable (Hoen et al, 2011).
Numerous attitudinal studies on how wind energy development is perceived at the community level have also emerged. Most of these surveys are either simple format or use extensive empirical stated preference method, look at off-shore wind turbines, and show mixed results. Ladenburg et al. (2007) in Denmark use choice experiments to estimate the willingness to pay for visual disamenities related to prospective offshore wind facilities. Their valuation scenario comprises the location of 720 offshore wind turbines in farms at different distance increments from the shore, relative to an 8 km (approximately 5 miles) baseline. They find that the average willingness to pay (WTP) increased proportionally as the wind facility distance from shore increased. In addition, their results reveal that WTP varies significantly depending on the age of respondents and their experience with offshore wind farms. This demonstrates the importance of understanding the demographics of homeowners and stakeholders.
The subjective nature of attitudinal studies makes them helpful when examining wind turbine effect on area stigma. One survey study by Goldman (2006) finds no evidence of area stigma from a survey of local residents conducted after the wind facilities were erected. Yet, another study finds limited evidence of these stigmas (Bond, 2008). The results from her surveys of public attitudes towards the construction of proposed wind farms indicate that although the overall respondents think of wind farms in positive terms with proximity as an important determinant factor, 38% of the respondents would pay 1% -9% less for their property due to the presence of a nearby wind farm. The large distance between properties to the wind farms in both study areas impose limitation on the significance of the results. Also, the author cautions generalization of these results since resident attitudes can be highly location-specific.
The correlation between how wind energy is perceived at the community level and wind energy development timeline is an important component in understanding its acceptance by communities. Homeowners' attitudes before and after wind energy facility construction are examined in two separate studies in the U.K. by Khatri (2004) and Warren et al. (2005). The findings from both studies suggest that when wind farm development is first announced, property prices may decline, but prices are likely to recover after the wind farms are in operation and communities learn more about the benefits of wind development. Similarly, Wolsink (1989) and Palmer (1997) examine public attitude in relation to wind energy development timeline. They find that local residents' attitude towards wind power is at its lowest during the planning stage, but nearly returns to the pre-announcement levels after the facilities are built. Studies by Devine-Wright (2004) and Thayer and Freeman (1987) show mixed results, but emphasize the importance of understanding local community perceptions.
Many studies reveal that wind energy facilities are predicted to negatively impact property values pre-construction (Khatri, 2004;Firestone, Kempton, and Krueger, 2007;Kielisch, 2009), but negative impact largely dissipates post construction (Sterzinger, Beck, and Kostiuk, 2003;Hoen, 2006;Poletti, 2007). Further study of wind turbine post construction effect is hindered by the fact that wind development is still in its early stage with a relatively small number of property sales transactions. Some studies look for supporting data from research on facilities that are "similar" to wind turbines.
Attitudinal studies of high voltage overhead transmission lines (HVOTL) show statistically significant negative impact on the price of properties in close proximity (Bond and Hopkin, 2000;Des Rosiers, 2002). Given the longer timeline of HVOTL facilities, these results can be used to supplement house sales transaction data in relation to wind turbines. HVOTL data may provide supporting information for potential post construction effects of wind turbines.

APPROACH OF THIS STUDY
This study uses an extensive data set focusing on residential communities in close proximity with single high capacity wind turbines. Much of the literature used largescale wind farms sited primarily in rural areas with relatively sparse property sales transaction observations. Wind turbines in Rhode Island are primarily located near residential areas virtually by necessity because of the state's small size and high population density. This provides the setting for more close proximity property sales transactions. This study analyzes house transaction data with respect to location and demographic specific nature in relation to twelve wind turbines in seven different towns in a single housing market. The single wind turbines in this study vary in size, but are industrial scale with capacity beyond typical residential electricity usage. The study investigates the potential negative impact of wind turbines on nearby property values, with residential demographics as control variables.
This study draws from the approach by Hoen et al. (2006;2007;2009;

STUDY AREAS
This study includes twelve single wind turbines sited in seven Rhode Island towns as shown in FIGURE 1. Minimum wind turbine energy capacity of 100 kilo-Watts (kW) is chosen for the study: four 100 kW, one 250 kW, one 275 kW, one 660 kW and five 1.5 Mega-Watts (MW). This is based on the assumption that wind turbines less than 100 kW would have qualitatively smaller, if any, impact.

Property Sales Transaction Data
This study uses the extensive RI house sales transaction dataset which comprise 69,768 Island Housing Price Index. The exterior condition is categorized in eleven levels from 1 being "unsound" to 11 "excellent", thus the mean value of 5.38 is between categories 5 ("average") and 6 ("average-good"). These are collapsed into three categories for the estimation model shown in the next section: below average, extcon1; average, extcon2; and above average, extcon3. Below average, extcon1 is used as the reference category.
Lot size in acres is used as a categorical variable to control house characteristics.
There is concern in the literature that not all property transactions are "arm-length" transactions. For example, some properties are sold within a family for a nominal price to satisfy legal requirements for ownership transfer. In these cases, sales price does not reflect true market value. As a consequence, transactions at anomalously low prices are commonly excluded from hedonic analysis. This study excluding observations with sales price less than $40,000, similar to Hoen et al., 2009Hoen et al., , 2011Walsh et al. 2011. Lot size in acres is used as one of the fixed effects. Only single family residential property transactions within 5 miles to the nearest wind turbines are used for the analysis.

Geographical Information System Data
GIS data from Rhode Island Geographic Information System (RIGIS) were used to calculate the distance between each house with a transaction and its closest wind turbine.
House locations were geocoded using ArcGIS geocoding tool with RIGIS address locator which is based on Emergency 9-1-1 structures. 4 Buffer proximity rings around the wind turbines combined with house locations were used to identify properties within each proximity increment of thirds of a mile (less than 0.33 mile; 0.34 to 0.66 mile; 0.67 to 1 mile), then 1 to 2, 2 to 3 and 3 to 5 miles. These are used for field data collections for view and % residential (or vista) variables. The geocoded addresses were also used for verifying and correcting the original longitudes and latitudes that came with the dataset.
RI 2010 Census information from RIGIS was used to verify Census tracts and blocks.
Using ArcMap NEAR tool, each transaction was assigned a unique distance (dist_mi) defined as the distance between the home and its nearest wind turbine at the time of sale.
The empirical modeling uses actual distance both as a continuous variable (dist_mi: distance between each property and its nearest wind turbine in miles.) and a vector of six discrete distance increments.

View and % Residential (or Vista) Variables from Field Data collection
The view of the wind turbine from a particular property is likely to be a significant factor in determining the effect on the property sales transaction prices. Although ArcTool software program has a viewshed tool, field data collection for view variable became necessary because the software could not take into account obstruction by nearby trees or neighboring houses (Hoen et al., 2009). Field data comprised of scenic vista in terms of percent residential versus industrial and view of the wind turbines were collected from site visits to each home in the post construction sample.
To ensure the consistency in rating, field data collection was conducted by the same These two qualitative measures were based on the categories defined by Hoen et al. (2009Hoen et al. ( , 2011 and modified for RI context of this study. Field data for three additional variables, "contrast" to supplement the vista and "viewnow" and "viewseason" were collected to differentiate the view by season, particularly tree foliage. However, the model selection process with various combinations of these variables, and the benefit of parsimony with variables for better modeling led only view and vista variables with modifications to be included in the final model. These potential variables are based on the presence of wind turbine view, but since only 0.2% of post construction properties have wind turbine view, this would not be expected to have a substantial impact. Since the degree of industrialization and residential development in neighborhoods showed more discernible correlation, they were incorporated into the vista variable as % residential.
There are a handful studies that conducted field data collection for view and vista variables to examine scenic vista stigma (Dent and Sims, 2007;Sims et al., 2008;Kielisch, 2009;Hoen et al., 2009Hoen et al., , 2011. They were drawn on the landscape-quality rating system developed by Buhyoff et al. (1994) and dubbed as Q-sort to differentiate the degree of sorting between two extreme ends (least vs. most). It categorizes by sorting the photographs of landscapes concerned in odd numbered groups, usually five groups as optimal: two for each extreme end and one for the middle reference point. There was 75-85% consistency categorizing photographs of the field landscapes between those personnel that were and were not involved with field data collection (Torres-Sibillea, For each view measure, three rubrics were used as reference for maximum consistency: scope of view from a property; the viewer angle of the view of wind turbine; and the degree of contrast between wind turbine and its surrounding objects. Combination of these rubrics is categorized by five levels with view 0 for no view as base category: view 1, minor view with less than 33% of the view scope; view 2, moderate view between 34% and 66% of view scope; view 3, substantial view between 67 and 80%; and view 4, extreme view above 80% of the view scope. The angle of the view is kept between 0 to 90 degrees of viewer's waist height as plane of zero and the viewer at standing position.
The view scope ranges from 0 to 180 degrees with viewer's left shoulder to her right shoulder. View categories are described in TABLE 10 and sample photographs are indexed in Appendix A.
A rating for the quality of the scenic vista (vista) from each property was also collected because view and vista are expected to be correlated. Properties with a premium vista are more likely to have a wide viewing angle for the view of wind turbines. In order to accommodate the high housing density inherent in a residential setting, the degree of urbanization in terms of percent commerce and municipal utility facilities in the neighborhood is integrated in the ranking. The lower number of vista indicates the higher percent of industrialization or commerce and less residential neighborhood. Five categories defined by Hoen et al. (2009Hoen et al. ( , 2011 (5) PREMIUM, 90% Residential. Vista categories are described in TABLE 11 and sample photographs are indexed in Appendix B. Observations from the vista (or % residential) field data collection indicate high correlation between percent residential in neighborhoods and the distance to coastline and high voltage overhead transmission lines (HVOTL). FIGURE 7A shows coastlines and HVOTL near wind turbines. Both coastline (ocean_mi) and high voltage overhead transmission lines (hvotl_mi) show more consistent correlation than % residential. They are also better quantitative indicators to which neighborhoods are residential or industrial compared to using % residential. The final estimation model includes both as control variables in lieu of including vista or % residential as regressors. Since view and vista are correlated, the vista component is accounted for in the view variable in the model (Hoen et al. 2009).

DEVELOPMENT OF HEDONIC MODELS
Hedonic pricing models have been used for characterizing the prices of competitively traded heterogeneous goods in numerous bundles. Hedonic consumer price index for appliances is a good example. Housing market is most commonly used for environmental hedonic models because of common spatial factors, such as location of houses and their surrounding environmental attributes. A given housing unit is best characterized as consisting of a bundle of attributes that in aggregation describe the structure itself, the land upon which it is built, and the relevant spatial characteristics. In addition, the hedonic approach can separate the internal property attributes (baths, bedrooms, square feet, etc.) from the public and private good attributes associated with location. For example, a hedonic pricing model identifies a premium paid for houses located near desirable environmental amenities, according to the premise that price is determined by both internal characteristics of the good being sold such as structural characteristics of a house and external factors such as environmental externalities (Freeman, 2003). In this study, the environmental amenity (proximity to wind turbine) is the non-market good characteristic, while the house is the market good.
A generic hedonic price function for housing market is a linear regression comprised of house sales price in most cases as a dependent variable and an array of numerous explanatory variables, grouped in the three main bundles as below: House Price = f (HC, NC, EA),

where HC is the set of structural house characteristics, NC is neighborhood characteristics and EA is environmental amenities.
House characteristics include number of bedrooms, number of floors, lot size, square footage of living area, number of fireplaces, type of heating system, type of exterior material and general condition among other things. Neighborhood characteristics usually include school quality, crime rate, and municipal services such as fire station, town police, and demographics of neighbors. Environmental amenities include presence or absence of a nearby park, farm, river or stream, open space, and lakes. Environmental disamenities might include air pollution, industrial facilities and landfills.
Hedonic price model is a revealed preference method through which an implicit marginal willingness to pay (MWTP) for a good can be estimated from consumers' selection behavior. It is a bid function within the supply and demand curves of a market in which the value of non-marginal characteristic changes within the context of the hedonic price function and the marginal bid function. 5 Hence, hedonic price model is a reduced form of equation that entails two stages. The first stage is a multivariable linear regression to observe the hedonic price function in a market to measure the slope of a characteristic of interest. The second stage uses this slope as a price of the characteristic to determine the demand for that characteristic. This is an important step because how much price changes when the characteristic of interest changes by one unit is the marginal price of that characteristic, which represents the marginal value, not the incremental value for a finite change in the characteristic. For example, if the marginal value of distance diminishes, using the estimate from the hedonic price function will overstate the WTP for a change in distance for a distance increase, and understate WTP for a distance decrease.

Limitations and Challenges of Hedonic Price Model (HPM)
In order to accurately estimate the effects of the wind turbines on home values with HPM, it is important to address three empirical issues embedded within the hedonic model: omitted variables, endogeneity, and spatial dependence or autocorrelation. There are many factors that co-determine the house price and many of these factors are unobservable, therefore not included for the analyses. If any of the unobserved factors are also correlated with included factors, then the resulting coefficient estimates will be biased due to omitted variables. Endogeneity bias arises when the values of the dependent and one or more independent variables are co-determined. For example, if the wind turbine is developed in the neighborhood of lower value properties and the presence of wind turbine lowers property values, we have endogeneity.
Endogeniety, spatial dependence or autocorrelation error and omitted variable bias can be corrected with fixed effects applied at a specified geographic range (Kuminoff, Parmeter, and Pope, 2010) and difference in differences estimation approach. At the most precise level, these fixed effects will occur at parcel level in terms of lot size in acres of the property. This study uses interactive terms along with local fixed effects to address these issues (Greenstone and Gayer, 2009). At a larger scale, fixed effects can be applied at Census block, block group, county or Census tract. This study uses Census Tract and region as spatial fixed effects for the difference-in-difference estimation.
Year-quarterly is used as temporal fixed effects.

Functional Forms and Model Selections
This study uses the logarithmic functional form. It is shown that econometric models for the equilibrium price function perform best when all variables are included in the model but that simpler functional form using a linear, log-linear specification performed best in the presence of omitted variable (Cropper et al., 1988). Logarithmic and semilogarithmic functional forms which represent the elasticity in percentage render easier interpretations. Linear and squared terms were tested for primary living area, age and lot size because theory and empirical results suggest nonlinearities in valuing these characteristics. However, for the purpose of determining the correlation of proximity to the nearest wind turbine, a linear form for the continuous distance variable is used, following the convention using a log-linear specification for log of sales price.

Following numerous trials with functional transformation, these two models [3] and [4]
are selected: ln (adj.Price ijt ) = λ t + α j + β 0 + ∑β 1 X ij + ∑β 2 N ij + where adjP ijt represents the price of the property i adjusted to the 2 nd quarter of 2010 RI housing price index in the fixed effects group j at time t; λ t represents the year-quarter fixed effects; α j denotes the census tract fixed effects; dist_mi ikt variable is the linear continuous distance in miles between a particular property and its nearest wind turbine.
diswt ikt is a vector of six discrete distance increments with a base of 3-5 mile increment. (dist_mi ikt )*(timeperiod ik ) is an interaction term between linear continuous distance in miles and timeline variables. (dist_mi it ) *(view ikt ) is an interaction term between continuous distance and a vector of view in four categories; the same holds for the discrete distance variable, diswt ikt . ocean_mi ikt is a linear continuous distance between a property and its nearest coastline in miles. hvotl_mi ikt is a linear continuous distance between a property and its nearest high voltage overhead transmission lines. ᶓ it is fixed effects error term.

DIFFERENCE-IN-DIFFERENCES ESTIMATION
The difference-in-differences (DID, henceforth) estimator has become a popular identification approach in recent literature (e.g., Bertrand et al, 2004), although it has a long history in analysis of variance. 6 It attempts to mimic random assignment of treatment and "comparison" (or control) sample by applying two-way fixed effects model: cross section and time fixed effects on pooled cross section data as of this study.
Accordingly, it requires four points of observations with two time periods (before and after the treatment) and two groups (treatment and control). One approach is DID without regression which simply takes the mean value of each group's outcome before and after treatment and then calculate the "DID" of the means. We can get the same result with DID with regression which allows us to add regression controls, if needed.
The regression framework that might pick up the effects of other factors that changed around the time of treatment is the attractive feature for its recent application popularity 6 Econometric Analysis of Cross Section and Panel Data, Second Edition by Jeffrey M. Wooldridge,p.148 because it uses a control group to "difference out" these confounding factors and isolate the treatment effect. Hence, the coefficient on the interaction term of two fixed effects in the regression framework is the treatment effect. And the overall effect is a coefficient sum of interaction term and its respective individual variables. In this study, the coefficient on the interaction term between logarithmic continuous distance between a property and its nearest wind turbine and time periods (pre announcement, post announcement and post construction) is the treatment effect -the impact by wind turbine on the property values of the nearest properties.

Development Anticipatory and Post Construction Impact by Wind Turbines
Wind turbine impact in terms of development timeline is explored with both linear continuous distance in  construction effect, if any existed, will most likely phase out as observed in other studies (Sterzinger, Beck, and Kostiuk, 2003;Khatri, 2004;Warren et al., 2005;Hoen, 2006;Poletti, 2007).

View and Proximity impact by Wind Turbines
TABLEs 7, 7A and 7B with interaction terms between viewshed and distance either in linear continuous (TABLE 7) or discrete increments (7B) show that the view impact by wind turbines is not significant. This may be due to high housing density which is generally associated with more obstructed views of wind turbine. In addition, since there were a small number of post construction observations, this warrants further study when more post construction observations are available.

CONCLUSION
This study shows no evidence that wind turbines have negative effect on property values.
If anything, the results show that distance to wind turbines has a statistically significant and positive impact during both pre announcement development and post construction in the magnitude of 2.45% and 1.96% respectively. The distance from the wind turbines has an even smaller effect after wind turbine construction: from 3.17% during pre announcement (PA), 2.45% post announcement (PAPC) to 1.96% after construction (PC). Overall, wind turbines have a positive impact on property price. This finding is consistent in both linear distance and discrete distance analyses. However, anomalous deviation in two discrete bands in discrete distance analysis suggests further investigation. DID estimates with discrete distance bands are based on the premise that property values in the 3-5 mile reference band are not affected by wind turbines. This is a common premise in multi-state wind farm hedonic studies. Since Rhode Island's numerous small towns are heterogeneous, further study using regional multiple comparisons may be considered.
Location-specific homeowner demographics influence the integration of new technology in society. It is important to include homeowners' concerns in decision making for wind energy development (Cowell et al., 2011). This study's use of residential demographics helps assess the impact of wind turbines on property sales prices, and provides a better understanding of community acceptance of wind energy facilities. It became clear through interviews and correspondence with RI town planners, and attendance at multiple stakeholders meetings, that transparency during the wind energy development process affects community acceptance. This is consistent with the discrete distance DID estimates and observation distribution by six different regions shown in TABLEs 4 and 5.
Towns with the most transparency have positive or less negative impact by wind turbines on property sales prices during both post announcement and post construction periods; towns with the least transparency have negative or more negative impact and attitude towards wind turbines.

Distance (relative to 3-5 miles)
2-3 miles                 The turbine is visible from this property but in very narrow view scope or angle due to many obstructions such as trees or neighboring properties. The distance between the property and the wind turbine is large.

View 2 Moderate view
The turbine is visible, but the view scope ranges from narrow to medium. There might be some obstructions but in much less degree and the distance between the property and the wind turbine is most likely a mile if not more.

View 3 Substantial view
The turbine is substantially visible from the property in a wide scope. A full profile of the turbine may be visible but not overbearing. The distance between the property and turbine is short, within a mile or less.

View 4 Extreme view
The turbine is dramatically visible to the extreme. This rating is reserved for sites that have unmistakably overwhelming presence of the wind turbine. The distance between the turbine and the property is very small. This table is a modified version of Hoen et al. study (2011) for the RI context of this study. These vistas are of boundary between urban and suburban. They include interesting views that can be enjoyed often but only in a narrow scope due to medium housing density.

Vista 3
Average vista 55% These vistas contain the properties relatively well maintained and convey a sense of comfortable community. They have a good balance of suburban appeal with the urban convenience.

Vista 4
Above Average vista 70% These vistas provide inviting views for people to enjoy in a medium to wide scope due to low housing density. Properties are relatively large and well maintained.

Vista 5 Premium vista 90%
These scenic vistas provide spectacular views to enjoy free or mostly free of man-made structures. The properties are large due to very low housing density, thus only a few neighboring properties. They convey a high potential for recreation.
This table is modified from Hoen et al. study (2011) for the RI context of this study.      ^Lot size is categorical variable to control house characteristics. PAPC stands for post announcement and pre construction; PC for post construction; the base category for time period is PA, pre announcement. Level of Significance codes: 0 '***' 0.001' 0.01 '*'0.05 '.'0.1 ' '1; 0.000 denotes the number less than 10e-6.