A hedonic analysis of the demand for and benefits of urban recreation parks

Land Use Policy 26 (2009) 975–983 Contents lists available at ScienceDirect Land Use Policy journal homepage: www.elsevier.com/locate/landusepol A hedonic analysis of the demand for and benefits of urban recreation parks Neelam C. Poudyal a , Donald G. Hodges a,∗ , Christopher D. Merrett b a Natural Resource Policy Center and Department of Forestry, Wildlife and Fisheries, The University of Tennessee, Knoxville, TN, United States b Illinois Institute of Rural Affairs, Western Illinois University, Macomb, IL, United States a r t i c l e i n f o a b s t r a c t Article history: Increasing population and urbanization in U.S. cities is not only contributing to the congestion in urban Received 18 February 2008 recreation parks but also is likely to exceed the capacity of these parks’ recreational and amenity benefits. Received in revised form In order to estimate the demand for and benefit of parks, we employed a typical hedonic model, which 25 November 2008 confirmed that the urban recreation park acres increase nearby property values. Two Step Clustering, Accepted 30 November 2008 which is capable of defining the optimum number of submarkets based on the data, was employed to define the submarkets within Roanoke, Virginia and to obtain enough implicit price points to further Keywords: estimate the demand for urban park acres in the second stage. Results from the second stage hedonic Urban recreation parks Demand estimation revealed that demand for urban park acres was inelastic in price and income; and the size Open space of the park was a substitute for living space and proximity to park. In addition, increasing the average Hedonic model size of parks by 20% from the current level increased the per household consumer surplus by $160. The estimated amenity benefits of urban recreation parks will be useful in urban landuse planning and open space preservation. © 2008 Elsevier Ltd. All rights reserved. Introduction the demand for urban parks and open space is likely to grow in the future with citizen awareness of environmental issues and Open space and public recreation lands enhance the economy demand for ecosystem services. Effective urban landuse planning and quality of life in cities by improving air quality, providing and supplying additional acreage of such parks will require a clear recreational opportunities, and enhancing aesthetic values, among understanding of their amenity values and demand in our soci- many other benefits (Nowak and McPherson, 1993). As cities grow, ety. Our current knowledge on the economic value of such parks is however, open spaces are paved over to make room for new limited, however, especially regarding the welfare effect to the com- buildings and roads needed to accommodate the increasing urban munity. One of the six major strategic goals of the National Research population. McPherson (2006) pointed out that one of the likely Plan for Urban Forestry 2005–2015, for example, is focused on prop- constraints in North American urban forestry is the overuse of erly estimating the economic benefits and real estate value added recreation parks in urban areas. As the population grows rapidly by the enhanced neighborhood quality from such parks (Clark et al., and prompts urban expansion, the remaining small forest patches 2007). This paper attempts to estimate the demand for recreation and open spaces in urban territories are being shared among an park acreage in urban neighborhoods using a two-stage hedonic increasing number of people (Kline, 2006). The size of open spaces framework. It will also assess how the supply of additional acreage not only affects the recreational potential and aesthetic values, but of land in such parks will increase the welfare in our society. also determines to some extent the quantity and quality of ecosys- Estimates of the economic values or amenity benefits of urban tem services provided to the nearby community (Metropolitan parks and public open spaces have emerged recently (Tyrvainen, Design Center, 2004). 1997; Tyrvainen and Miettinen, 2000; Bolitzer and Netusil, 2000; Public support for open space protection has increased sub- Lutzenhiser and Netusil, 2001; Geoghegan, 2002; Hobden et al., stantially in recent years. For instance, public referenda on open 2004; Salazar and Menendez, 2007). A review of the open space space have passed in at least 39 states in the U.S. since 1999 and urban park literature by McConnell and Walls (2005) analyzed (Trust for Public Land, 2005). In addition to population growth, the results from studies focusing on different kinds of open spaces. Most of the studies focused on either the distance to urban parks or the proportion of open space in some defined level of neighborhood ∗ Corresponding author at: The University of Tennessee, Natural Resource Policy to measure their amenity values in dollar terms. With some notable Center, 274 Ellington PSB, Knoxville, TN 37996-45463, United States. exceptions (e.g., Bolitzer and Netusil, 2000), most have ignored the E-mail address: dhodges2@utk.edu (D.G. Hodges). acreage effect of urban parks on property values. 0264-8377/$ – see front matter © 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.landusepol.2008.11.008 976 N.C. Poudyal et al. / Land Use Policy 26 (2009) 975–983 Previous research revealed that the price of a house increases however, suggest some appropriate substitute/complementary with its proximity to nearby parks (Tyrvainen, 1997; Tyrvainen and measures for environmental amenities (e.g., Boyle et al., 1999). Miettinen, 2000; Thorsnes, 2002). Similarly, other studies revealed Although application of second stage demand was fairly common that increasing the size of urban parks increases the housing values in earlier housing studies (Palmquist, 1984; Allen et al., 1995), nearby (Tyrvainen, 1997). Salazar and Menendez (2007) estimated researchers have recently begun employing this in demand anal- the willingness to pay for proximity to planned urban parks in ysis of environmental amenities (Boyle et al., 1999; Brasington and Spain. Likewise, Bolitzer and Netusil (2000) considered the size Hite, 2005). of different natural areas on property values in Portland, Oregon, Most of the previous studies relied on data from multiple mar- and noted that public parks significantly increase housing prices in kets or metropolitan areas to identity the demand (Taylor, 2003), the neighborhood but private parks do not. Lutzenhiser and Netusil and ignored the existence of submarkets within a single city (2001) concluded that the size of natural areas and parks has the (Palmquist, 2005). Recent literature on housing and real estate eco- largest effect on property prices compared to any other kind of open nomics indicates that segmented submarkets exist even within a space. single city, where the dwellings within a submarket serve as sub- Anderson and West (2006) and Morancho (2003) reported that stitutes (Bourassa et al., 1999). Recent studies utilizing hedonic the size of urban parks or green areas did not have a significant methods have utilized statistical clustering techniques to identify amenity effect. In a recent hedonic study, Mansfield et al. (2005) submarkets (Lipscomb and Farmer, 2005, Day et al., 2007; Cho et concluded that trees in the lot or immediate neighborhood could al., 2008). However, estimating the second stage demand function serve as substitutes for living near large woodlots. Despite the sub- to evaluate the urban parks or open space has not been a focus of stantial amount of research, little information is available on how previous studies. increasing the area of urban recreational parks will be capitalized Mahan et al. (2000) estimated the demand for wetland acres in the real estate market and local property tax base. Understand- using single city data from Portland, Oregon, but failed to yield ing how economic welfare changes by supplying more recreational theoretically consistent results, probably because they relied on open space in the urban landscape and how residents respond to arbitrarily defined submarkets for demand identification. This such policies is important before taking initiatives on open space study aims to improve upon this issue by properly identifying the protection and smart park design. submarket within a city using a statistical clustering technique, Another limitation of previous studies is that they relied on the and demonstrating that the hedonic function differs significantly implicit price from the first stage hedonic regression to explain the among those submarkets. If the submarkets differ in hedonic func- economic benefit from living close to open spaces. Policy changes tion, then observations from those submarkets should be similar to are often non-marginal, however, and those implicit prices can- observations from multiple cities and demand could be identified not measure such benefits ex-ante (Taylor, 2003). Therefore, proper (Taylor, 2003). By estimating the second stage demand, this study estimation of demand requires second stage estimation (Rosen, also evaluates the welfare effects from non-marginal changes in 1974), which combines the observed quantity of urban park acreage park size. Findings from this study could be useful in urban landuse with the implicit prices from the first stage hedonic regression to planning and understanding the public demand for and welfare estimate the demand function. For the non-marginal changes, the effects of recreational open space projects. identification problem arises in estimating the demand (see Taylor, 2003). Recent work on identification in hedonic models by Ekeland Materials and methods et al. (2002, 2004) discusses this issue in detail. The identification problem is addressed either by estimating the hedonic regression Study area from multiple markets (e.g., Palmquist, 1984; Boyle et al., 1999; Brasington and Hite, 2005) or by using the marginal rate of sub- The study was conducted with data from the City of Roanoke, stitution (MRS) approach (e.g., Chattopadhaya, 1999). Due to some Virginia. Roanoke was chosen because: (1) it is the largest and questionable assumptions in the MRS approach (Freeman, 1993; fastest growing urban area in southwestern Virginia, (2) it has been Boyle et al., 1999), and the robustness in the source of identification listed among the top 3 most livable small cities in the nation for its in the multiple market approach (Ekeland et al., 2002), the second amenity attractions including urban parks (PLC, 2004), and (3) it approach is preferred. The idea behind the multiple markets is that contains several urban recreation parks of varying size, uniformly the marginal implicit price of the attribute of interest varies with distributed throughout the city and hence provides an ideal place the estimated parameters across those markets. for the study of urban park benefits. The city area includes 46 differ- This study attempted to fill the gaps discussed above in hedo- ent urban parks. In addition to green open space and urban trees, nic valuation of urban recreation park acreage. In particular, this these parks are supplemented with additional man-made attrac- study evaluated the amenity value of urban recreation parks and tions such as greenways and playgrounds. estimated the demand for park acreage based on a second-stage hedonic demand that employed the implicit price of park acreage from various submarkets within an urban area. Second stage The data demand estimation is challenging because it involves additional data and estimation requirements beyond the first stage hedonic Data used in this study came from a variety of sources. Housing regression – which has been the common technique for evaluating prices and their structural characteristics were obtained from the amenities in most of the existing hedonic studies. This is because Geographic Information System (GIS) database of the real estate the identification of the demand curve of an attribute (e.g., urban department at City of Roanoke, Virginia. A total of 11,334 single- park acreage) requires variation in its implicit price (Boyle et al., family houses were sold between 1997 and 2006. However, 209 of 1999). Since we estimate one point on the demand function with those did not have complete information or did not seem to involve a single hedonic model, data from different markets are needed arms-length transactions, and were omitted from the analysis. This to estimate other points, controlling for differences in preferences resulted in a final dataset consisting of 11,125 houses with com- across markets. Another difficulty with the second stage estimation plete information. All housing prices were adjusted to year 2000 is identifying appropriate measures of substitute and comple- dollars to control for real estate market fluctuations in the city, and mentary goods to complete the demand function. Recent studies, to make them compatible with the neighborhood data of Census  N.C. Poudyal et al. / Land Use Policy 26 (2009) 975–983 977 Table 1 Definition and descriptive statistics of variables. Variables Definition Mean Standard deviation Structural and housing variables Living area Square footage of the living area in house 1,388.20 608.54 Brick exterior Dummy variable, 1 if the exterior is brick, 0 otherwise 0.51 0.49 Bedrooms Number of bedrooms in house 3.01 0.42 Central AC Dummy variable, 1 if the house has a central AC, 0 otherwise 0.69 0.45 Garage Dummy variable, 1 if the house has a garage, 0 otherwise 0.24 0.42 Masonry fireplace Dummy variable, 1 if the house has a masonry fireplace, 0 otherwise 0.46 0.49 Parcel Square footage of the parcel 11,872.71 13,652.44 Age Age of house in year 2006 54.84 27.12 Season Dummy variable, 1 if the house was sold in months 4–9, 0 otherwise 0.55 0.49 Stories Number of stories in the house 1.25 0.57 Enclosed porch Dummy variable, 1 if the house has an enclosed porch, 0 otherwise 0.17 0.38 Hip roof Dummy variable, 1 if the house has a hip roof structure, 0 otherwise 0.13 0.34 Neighborhood and urban recreation park variables Population density People per square miles in the census block group 3,088.89 1,668.02 Poverty Percentage of residents under poverty level in the census block group 12.50 10.71 Vacancy rate Proportions of vacant houses in the census block group 0.06 0.03 Black Percentage of black population in the census block group 20.85 28.21 Median age Median age of the residents in the census block group 38.51 5.65 Median household income Median household income of the residents in the census block group 35,174.10 15,500.05 College degree Percentage of residents with college degree in the census block group 9.17 8.17 School Distance in feet from the house to the nearest school 3,128.55 2,052.87 Public bus Distance in feet from the house to the nearest public bus route 1,239.16 1,450.83 Airport Distance in feet from the house to the regional airport 14,338.34 7,490.71 CBD Distance in feet from the house to the central business district 13,307.11 5,542.85 Railroad Distance in feet from the house to the nearest railroad track 3,938.55 3,145.32 River Distance in feet from the house to the nearest river or stream side 2,704.48 2,042.04 Park size Size in square footage of the nearest urban recreation park 2,527,081.00 5,503,957.00 Park proximity Distance in feet from the house to the nearest urban recreation park 2,281.72 2,244.95 Year 2000. For this, we used the annual housing price index for estimation. A White test of heteroscedasticity was used to test the City of Roanoke, which was obtained from the Office of Federal for the homoscedasticity of the error term. If heteroscedastic- Housing Enterprise Oversight (OFHEO, 2006). Since the month in ity was present, consistent estimates of the standard errors were which sales occurred was also recorded along with the transaction obtained using White’s approach of robust standard error (White, price, the month of sales was used to create a season of sale dummy 1980). Presence of multicollinearity was tested using the Variance variable. Inflation Factor (VIF), which is a scaled version of the multiple cor- The GIS files of spatial location of parcels, regional airport, public relation coefficients among independent variables in the model bus routes, schools, railroad, rivers, urban parks, central business (Greene, 2003). We used some of the independent variables in districts were obtained from the city as well. Distances from each logarithmic or square form. Taylor (2003, pp. 353) suggested that house to those features were computed in ArcGIS 9.2. The neighbor- functional forms in hedonic models can often have some or all of hood data on socioeconomic information were obtained from the the independent variables transformed because marginal price may census block group1 level data of US Census Bureau data of 2000 not be constant for all characteristics. (US Census, 2000). Descriptive statistics of the variables used are As expressed in the hedonic function above, three sets of presented in Table 1. explanatory variables were used. These included a set of structural variables of the house, a set of neighborhood characteristics, and a Hedonic model set of variables explaining the attributes of the nearest urban park. Key structural variables included size of the living area, number This study used a typical hedonic equation of housing price in a of stories, age of house, and size of the parcel on which the house semi-logarithmic form as shown in Eq. (1). is located. In addition, dummy variables were included to indicate    whether or not the house has a brick exterior, central air condition- ln Pi = ˇ0 + ˇj Sij + ˇk Nik + ˇl Uil + εi (1) ing (AC), and hip roof design. Other dummy variables were included to capture the presence of a masonry fireplace, an enclosed porch, where lnPi is the natural log of real sales price of the ith house, or a garage. A seasonal dummy was also included to control for Sij represents the jth structural variable, Nik is the measure of the price differential in houses sold in the spring-summer and fall- the kth neighborhood characteristic, and Uil represents the lth winter seasons. attribute of urban park. Similarly, ˇ0 , ˇj , ˇk , ˇl , represent the Variables capturing the neighborhood characteristics included corresponding parameters to be estimated, whereas εi captures population density, percentage of population below poverty level, the stochastic error term. Although, it is common to estimate the and vacancy rate of houses. Population density captured the rel- above model with ordinary least squares (OLS), estimating Eq. (1) ative congestion and level of development in the neighborhood. using OLS assumes that the variance of the error term is con- Percentage of population below poverty level reflected the eco- stant (i.e., homoscedastic). But if this assumption is violated, the nomic condition and prosperity of the neighborhood. The vacancy error terms will be heteroscedastic, thereby undermining model rate captured the housing occupancy and residential consumption rate in the neighborhood. Distance variables were included to capture the proximity of 1 As of 2000 census, there are 82 different census block groups within the city of the house to several amenities and disamenities in the neighbor- Roanoke. hood. Those included distance to the central business district (CBD), 978 N.C. Poudyal et al. / Land Use Policy 26 (2009) 975–983 distance to airport, distance to nearest school, distance to pub- teristics entering the hedonic function. The first step of the Two Step lic bus route, distance to railway tracks, and distance to riverside. Cluster method begins with pre-clustering observations for individ- These distance variables were expected to control for the locational ual houses by constructing a likelihood distance measure function. effects and bring useful information in submarket identification, A matrix containing distances between all pairs of pre-clustered which will be discussed later. Distance to CBD measures the prox- observations is created. In the second step, these pre-clustered imity to employment and business hub in the area. Goodman and groups of original observations are treated as individual observa- Thibodeau (1998) mentioned that the quality of public education tions and re-grouped by selecting the optimal number of clusters is the major predictor of neighborhood quality and can explain the using either the Bayesian Information Criterion (BIC) or the Akaike housing segmentation. By the same token, the distance to school Information Criterion (AIC). Because the first step groups a large was also included. Other distance variables mentioned above were number of original observations into a much smaller number of expected to control for other amenities and disamenities in the pre-clusters, the second step uses an agglomerative hierarchical locality. clustering to re-group the pre-clusters (Green and Salkind, 2003). The size of the nearest urban recreation park, the focus variable Following McGarigal et al. (2000), and Strong and Jacobson (2005), in the study, was included as well. In addition, the distance from this study employed Two Step Clustering techniques for market the house to the nearest urban park was also included to account segmentation. Recent studies have also used alternative methods to for the accessibility of the park from the house. A positive and sig- define the optimum number of submarkets. For example, Lipscomb nificant effect of the size of urban park was expected. That means and Farmer (2005) used an endogenous sorting process, which first the larger (smaller) the size of the park, the larger (smaller) the uses principal component analysis (PCA) to extract orthogonal fac- sales price of nearby houses. In addition, distance to the park was tors crossing a threshold Eigen value; and then employs an iterative expected to have a significant and negative effect. Detailed defini- seemingly unrelated regression approach to identify optimal clus- tions of the variables used in the model are presented in Table 1. ters of households, based on maximization of a G-statistic. This While developing a robust hedonic model to yield theoretically con- technique is similar to Two Step Clustering in the sense that both of sistent results, we included as many structural and neighborhood these rely on maximizing some mathematical function while classi- variables as possible in the hedonic model in order to properly tease fying observations into an optimal number of homogenous groups. out the effect of our focus variable (i.e. park size) on property price. Another segmentation technique recently suggested by Paez et al. Lipscomb and Farmer (2005), however, used separate variables to (2008) uses a moving window approach to optimally define hous- segment the market through an endogenous sorting process, and ing submarkets. to estimate the hedonic models. Our use of variables is consistent We considered three factors in identifying submarkets. First, with the approach of a number of similar studies mentioned earlier houses within a submarket are close substitutes for one another, (e.g., Day, 2003; Chen et al., 2009; Cho et al., 2008), and was nec- but poor substitutes for houses in other submarkets (Bourassa et essary to define the submarkets that are homogenous in as many al., 1999; Grisby et al., 1987). So it is important that the houses attributes as possible and to estimate the coefficient of the focus classified as members of a submarket be similar in their proper- variable accurately. ties. Second, the distribution of household’s tastes and preferences In order to identify the demand, the above-discussed hedonic should be stable or the same across markets but price functions model was estimated for different submarkets (to be discussed in should be different (Ekeland et al., 2002). Therefore, it is important the next section) within the city. The partial derivative of the esti- to control for the sociodemographic and economic characteristics mated hedonic equation (Eq. (1)) with respect to the attribute of of the households. Third, the ‘location’ has been argued to be the interest (i.e., size of urban park) was used to calculate the implicit most important factor in existence of submarkets (Bourassa et al., price of the park per acre. The implicit prices of such attributes 2003). So neighborhood characteristics and locational references from each submarket were then combined to estimate the demand (e.g., distances to various dis/amenities) should be considered in the second stage (to be discussed in Section ‘Demand for urban when delineating submarkets. Schnare and Struyk (1976, pp. 150) recreation parks’). viewed market segmentation as a function of structural attributes, neighborhood attributes, or a combination of both. We followed Market segmentation prior studies (Bourassa et al., 1999; Bourassa et al., 2003; Day, 2003; Chen et al., 2009; Cho et al., 2008) in using the variables Following Lipscomb and Farmer (2005), Lipscomb (2006), Day of a typical hedonic model in clustering and defining submar- et al. (2007), Cho et al. (2008), we first identified the submarkets kets. All structural properties, neighborhood characteristics, and and then estimated a separate hedonic price function for each sub- spatial distance variables (Table 1) were used in defining submar- market. A commonly used statistical technique to identify market kets, because we had to classify the houses in such a way that segmentation within a city is a k-means clustering (Bourassa et al., the resulting submarkets are homogenous in as many attributes 1999; Day, 2003; Chen et al., 2009; Cho et al., 2008), in which the as possible. housing and neighborhood variables forming the hedonic function Even though the clustering technique groups the properties of are used to group the houses into different clusters. Houses belong- similar characteristics into clusters, we do not know whether those ing to a particular cluster share similar structural and neighborhood clusters represent the submarkets in an economic sense. Therefore, characteristics and are substitutes for each other (Bourassa et al., ANOVA tests were performed to compare the statistical difference 1999). Hence, these clusters are commonly referred to as submar- among the submarket properties, based on sample means and vari- kets (Bourassa et al., 1999; Bourassa et al., 2003; Day, 2003; Cho ances (Moore and McCabe, 2003). Following Allen et al. (1995) and et al., 2008). However, use of k-means clustering is challenging Day (2003), a series of Chow tests were used to confirm the exis- because it needs a priori information on the number of possible tence of submarkets by determining if the hedonic price functions clusters or submarkets before segmentation can occur. Moreover, among those clusters differed significantly. k-means clustering cannot identify the optimal number of clusters or submarkets (Bourassa et al., 1999). Demand for urban recreation parks A recently developed clustering technique called Two Step Clus- tering (McGarigal et al., 2000) is capable of identifying the optimal The implicit prices of different attributes estimated from sepa- number of clusters based on the housing and neighborhood charac- rate hedonic regressions for individual submarkets were combined  N.C. Poudyal et al. / Land Use Policy 26 (2009) 975–983 979 to estimate a citywide-pooled model. Following Palmquist (1984), Table 2 Estimates from the hedonic regression. Boyle et al. (1999), Brasington and Hite (2005); a semi-log demand model was estimated. Variables Coefficients (standard errors) VIF Constant 4.229** (0.213) – ln Qui = Pui + Psc i + ıZi + i (2) Structural and housing variables ln(Living area) 0.657** (0.020) 3.41 Brick exterior 0.087** (0.008) 1.48 where ln Qui represents the natural log of acres of urban recreation Bedrooms 0.019** (0.006) 1.56 Central AC 0.183** (0.009) 1.33 park that is closest to the ith house, Pui is the implicit price of the i is a vector of implicit price for substitutes Garage 0.059** (0.009) 1.20 park acres. Likewise, Psc Masonry fireplace 0.139** (0.009) 1.73 and/or complements and Zi is a vector of exogenous demographic ln(Parcel) 0.041** (0.010) 1.57 and economic factors that are related to tastes and preference of the Age −0.003** (0.000) 16.53 residents and are expected to influence the demand. Similarly, , , Age squared −0.000 (0.000) 17.73 Season 0.047** (0.007) 1.00 and ı are the parameters to be estimated. While a typical demand Stories −0.046** (0.008) 1.92 model should consist of variables that control for substitutes and Enclosed porch −0.036** (0.010) 1.11 complements of a good under consideration, researchers have used Hip roof −0.014 (0.013) 1.24 subjective judgments in selecting such variables. Implicit prices Neighborhood and urban recreation park variables of living area and some attributes of the amenities under study ln(Population density) 0.013* (0.006) 1.48 have commonly been used to control for substitutes/complements Poverty −0.012** (0.000) 1.89 (Boyle et al., 1999). By the same token, implicit prices of living areas Vacancy rate −0.837** (0.150) 1.69 i . The ln(School) −0.033** (0.005) 1.31 and proximity to the park (in miles) were included in vector Psc ln(Public bus) 0.015** (0.004) 1.59 implicit price for park acreage and living area were log transformed ln(Airport) 0.066** (0.005) 1.75 for two obvious reasons. First, these variables exhibited substantial ln(CBD) 0.137** (0.015) 3.22 variation, and log transformation reduces the heteroscedasticity. ln(Railroad) −0.026** (0.005) 2.05 Second, the estimated coefficients can be directly interpreted as ln(River) 0.021** (0.004) 1.53 ln(Park size) 0.030** (0.002) 1.16 price elasticity. ln(Park proximity) −0.016** (0.004) 1.54 The exogenous demand shifters in Zi were taken from the corre- Adj. R2 0.64 sponding census block group information. As these socioeconomic F-statistic 563.98** characteristics were measured at the census block group scale, Number of observations 11,125 those might not be perfectly exogenous. However, for the hedo- Note: Dependent variable is ln(real housing price); the standard errors are white’s nic analysis they are valid demand shifters and have been used in robust standard error; ** and * denote the significance of parameter at 1% and 5% several hedonic demand models (Mahan et al., 2000; Brasington level, respectively. and Hite, 2005). These included the income of the home purchaser as measured by the natural logarithm of the median income of the households, tastes and preferences of the homebuyer as measured Results by median age of residents; race defined as the percentage of the population in the census block that are African–American; edu- As stated in the ‘Introduction’, the ultimate goal of this paper was cation as measured by the percentage of residents with a college to explore how economic welfare changes by supplying more recre- degree. ational open spaces in the urban landscape. Understanding how Similarly, following Clark and Cosgrove (1990), squares of the residents respond to such policies is important before taking some of these variables including median age and percentage of initiatives on open space protection and smart park designing. This African–American population were also included. Palmquist (1984) was accomplished by utilizing a second-stage hedonic demand that and Boyle et al. (1999) argued that the implicit price of the park employed the implicit price of park estimated from the first stage size and the income of the home purchaser are endogenous to hedonic model for various submarkets. The results of this effort are the demand function (i.e., Eq. (2)). So the endogeneity of implicit described below. price and median household income was addressed using instru- mental variables in a Two Stage Least Square approach. Following First stage hedonic model Palmquist (1984) the per capita income and the unemployment rate were chosen as instruments. A White test of heteroscedasticity rejected the null hypothesis We also estimated a welfare effect due to the change in acres of homoscedasticity at the 1% level in citywide hedonic regression of urban recreation park. Results from the estimation of Eq. (2) (Chi-square statistic = 1434.61, critical value of 135.81). Computed were utilized to estimate the marginal implicit price of the park values of VIF did not exceed the threshold value of 10 except for acres at its current mean level and at a 20% higher level. Using the age and square of age (Table 2). However, those variables were not MAPLE 9 program, consumer surplus was then calculated by first omitted, because using the square of age is a common practice in the integrating the demand function2 with respect to predicted prices hedonic literature to account for declining prices based on the age and then evaluating the integral at these two levels of prices. The of the house. Results from the citywide hedonic regression along other variables were kept constant at their mean level. with White’s robust standard error are presented in Table 2. Con- ventional adjusted R2 (0.64) reveals a relatively good fit of the data into the specified model. Coefficients on most of the variables (22 out of 24) were statistically significant at the 1% or better level and most of them had the expected signs.  2p2Fora −b logarithmic demand function lnX = a − b, lnP, the difference in integral Most of the structural variables including square footage of liv- e P dPevaluated at two price levels i.e., p1 and p2 were used to estimate the p1 change in consumer surplus for a household. X represents park acres, P represents ing area, exterior brick dummy, number of bedrooms, central AC predicted prices at a different level of X, whereas a and b are the Grand Constant and dummy, garage dummy, and masonry fireplace dummy were pos- the elasticity of price, respectively. itively related and significant at the 1% level, as expected. Not 980 N.C. Poudyal et al. / Land Use Policy 26 (2009) 975–983 Table 3 Means for property characteristics by submarket and ANOVA results of difference among submarket means. Variables Submarkets F-statistic 1 2 3 4 Living area 1,180.12 1,344.24 2,299.04 1,192.92 2032.13† Brick exterior 0.25 0.94 0.67 0.13 3901.97† Bedrooms 2.95 3.03 3.52 2.82 356.00† Central AC 0.91 0.70 0.91 0.45 736.43† Garage 0.19 0.29 0.48 0.11 307.29† Masonry fireplace 0.22 0.68 0.95 0.17 1987.71† Parcel 13,082.29 10,544.25 22,908.94 7,926.33 493.62† Age 30.75 60.77 47.36 67.30 1281.29† Season 0.54 0.56 0.59 0.52 8.67† Stories 0.90 1.31 1.47 1.33 439.69† Enclosed porch 0.06 0.21 0.15 0.21 94.83† Hip roof 0.02 0.13 0.16 0.19 131.44† Population density 1,928.16 3,615.63 2,021.80 3,705.74 1120.10† Poverty 8.89 8.27 4.65 22.75 2891.05† Vacancy rate 0.04 0.05 0.04 0.08 838.79† Black 17.75 15.10 3.87 36.20 696.76† Median age 37.09 38.92 46.22 35.78 1854.25† Median household income 38,104.02 35,005.10 56,414.81 24,587.54 2499.90† College degree 6.53 10.57 22.93 3.63 4553.29† School 5,322.36 2,603.99 2,683.91 2,446.23 1593.05† Public bus 2,659.48 760.97 1,874.33 563.53 1827.11† Airport 14,534.27 12,282.88 23,159.04 12,770.09 1062.67† CBD 18,209.65 13,203.89 16,215.55 8,993.99 2373.99† Railroad 6,060.88 4,026.88 3,525.39 2,623.46 666.24† River 2,080.60 2,800.58 2,520.70 3,085.50 123.40† Park size 3,865,281.00 1,115,257.00 6,893,779.00 1,365,865.00 563.79† Park proximity 4,106.21 1,537.90 3,790.23 1,266.15 1582.28† House price 89,739.22 90,773.50 215,755.30 53,156.13 2468.18† † Significant at the 1% level. surprisingly, these are consistent with existing hedonic studies. in square footage of the urban park in the neighborhood increased Likewise, the coefficient on square feet of the parcel and age of the the real sales price of the house by 0.03%, ceteris paribus. Taking house were significant at 1% level and were respectively positively the mean price of housing in the area ($95,133.99), the estimated and negatively related to house price. Some of the other variables elasticity indicates a marginal implicit price of $0.79, or roughly such as number of stories and enclosed porch dummy were neg- an $80 increase in price of nearby houses in response to a 100 ft.2 atively related, counter to our expectation. Similarly, among the increase in the size of the park. This finding confirms that the size neighborhood variables, population density was positively related of the nearby urban recreation parks has a small but significant and with house price, whereas the percentages of population in poverty positive relationship to property price. and vacancy rate were negatively related to house price at the 1% In the same way, distance from the house to the nearest park level. Distance from the house to the nearest school was negatively was negatively related to house price at the 1% level. This is con- related to house price, whereas the distances from the house to the sistent with the findings of Bolitzer and Netusil (2000), Tyrvainen closest public bus route and regional airport were positively related and Miettinen (2000), Morancho (2003), Tajima (2003), Anderson to the house price, all of them being significant at 1% level. Further, and West (2006), etc. The elasticity indicates that a 1% decrease the coefficient on the distance from house to the CBD was posi- in distance from the house to these parks increases the real price tively related and significant at the 1% level, and is quite consistent of the house by 0.016%, ceteris paribus. Combining the mean real with the economic theory because the houses farther away from the housing value and initial distance of 1 mile, the marginal implicit downtown might have a higher demand for residential purposes price3 would be $0.288, suggesting a $288 increase in house price than their downtown counterparts due to sound and quality of life. by moving the house 1000 ft. closer to an urban recreation park. This merely confirms the concentric zone model of urban develop- ment proposed by sociologists Robert Park and Ernest Burgess in Second stage demand model their 1925 study of Chicago. On the other hand, the observed effect of distance from house to the railroad was not palpable; the posi- Two step cluster analyses yielded four distinct submarkets tive effect of distance from the house to river might be attributed within the city of Roanoke. The number of submarkets identified to the fact that this variable did not distinguish between rivers and here is comparable with Cho et al. (2008), who found five dis- small streams or creeks, which might increase the risk of erosion tinct submarkets within the city of Knoxville, Tennessee, which and landslide (MacDonald et al., 1990). is similar in size and geographically close to Roanoke. Differences More importantly, the variables associated with the urban among characteristics of the property by submarket are reported park, which also is the focus of this study, were significant and had expected signs. The size of the nearest urban recreation park was significant at the 1% level and was positively related 3 We also tried an alternative model that included an interaction term of dis- to house price, corroborating the findings of previous studies (e.g., Tyrvainen, 1997; Bolitzer and Netusil, 2000; Lutzenhiser and tance to park with the size of park. However, the implicit price of distance in that model was only within 1.5% below the price calculated here. Nevertheless, a posi- Netusil, 2001). Since the sizes of parks measured in square feet tive and significant coefficient of interaction term suggested that people might place were log-transformed, as was the dependent variable, the coeffi- more value on larger parks located farther away than smaller parks nearby, possibly cients are interpreted as elasticity. It revealed that a 1% increase because of more opportunities for pets and human recreation in larger parks.  N.C. Poudyal et al. / Land Use Policy 26 (2009) 975–983 981 Table 4 Chow test statistics for equivalency of hedonic price function in submarkets. Submarket 1 2 3 1 2 11.34*** 3 13.74*** 22.03*** 4 5.46*** 15.43*** 20.49*** *** Denotes the significance of F-statistic at 1% level. Table 5 Demand estimation from Two Stage Least Square (2SLS). Variables Coefficients (standard errors) Fig. 1. Demand curve for acres of urban recreation parks. Constant −0.974** (0.181) ln(Price of park acre) −0.843** (0.008) ln(Price of living area) 0.636** (0.010) ticity using White’s method5 ). The own implicit price, price of Price of park proximity mileage 0.002** (0.000) ln(Median household income) 0.431** (0.022) substitutes, and exogenous demand shifters explained most of the Median age of residents 0.050** (0.006) variation in quantity demanded. In view of the goodness of fit of Square of median age −0.000* (0.000) the model and theoretically consistent estimates, it is reasonable Black −0.009* (0.000) to argue that the proper segmentation of submarkets in fact allows Square of black 0.000 (0.000) one to identify and estimate the demand function in the hedonic College degree −0.002** (0.000) framework, even with observations from a single city. In this regard, Adj. R2 0.94 we improved upon the method used for demand identification in a F-statistic 4239.62** two-stage hedonic method. Number of observations 11,125 As expected, the own price was negatively related to the acres Note: Dependent variable is ln(urban recreation park acres); the standard errors are of park demanded and was significant at the 1% level. Since both white’s robust standard errors; ** and * denote the significance of parameter at 1% and 10% level, respectively. the dependent variable and the price of park acres are in log form, the reported parameter is interpreted as elasticity. The elasticity value of −0.84 reveals that a 1% increase in the implicit price of in Table 3. The P-values of F-statistic in ANOVA tests indicated that park acres decreased the demand by 0.84%, ceteris paribus. This the property characteristics varied significantly among submarkets, indicates that the demand for urban recreation park acres was suggesting a clear distinction between those submarkets. In addi- inelastic, corroborating Kristrom and Riera (1996). Following Boyle tion, a series of Chow tests showed that F-statistics were significant et al. (1999), the demand function was evaluated for various lev- at 1% level for each submarket combination (Table 4), rejecting the els of park size and using a grand constant which contained all null hypothesis of equivalency of hedonic price functions between variables other than own price in their current mean values. Fig. 1 those submarkets. These evidences confirmed that Roanoke con- reveals a clear downward sloping demand curve for park acres with tains four distinct submarkets, which were utilized to identify the respect to its hedonic price change, indicating that urban residents demand. Like our statistical tests proved, the difference in iden- prefer larger parks to smaller ones, but they possess a diminishing tified submarkets in the area appears to have economic meaning marginal willingness to pay for the extra acreage. as well. Submarket 1 appears to contain medium-valued houses The coefficient of implicit prices of living area was positively located farther away from the downtown in a less dense area and and significantly related to demand for park acreage at the 1% level, is a relatively greater distance from the parks, public transporta- confirming that the house was a substitute for the size of nearby tion and schools. Submarket 2 appears to contain medium-valued parks. As revealed by the cross price elasticity of 0.63, the resi- houses made of brick walls, in densely populated areas with recre- dents demand larger park areas, if the cost of living space is high, ation parks and public transportation nearby. Likewise, submarket ceteris paribus. Earlier, Thorsnes (2002) found larger lot sizes to 3 includes higher-valued suburban houses occupied by high income be to some degree a substitute for open space in forest preserves. and well-educated people, whereas submarket 4 appears to be The cross price elasticity in case of distance to park was highly poorer housing in inner city areas. As in Day (2003), Lipscomb inelastic but significant and confirms that the proximity to park (2006), and Cho et al. (2008), the submarkets in our case were not is a substitute for size. A similar study by Mansfield et al. (2005) necessarily contiguous. However, the members of each submarket revealed that trees on a parcel could serve as substitutes for liv- formed apparent geographical clusters because we included a spa- ing near large forest blocks. These results might have important tial locational factor (proximity to various amenities.) in the cluster implications for real estate design and landuse planning. The exoge- identification process. Separate hedonic estimations4 were used to nous shifters of demand were also significant and had the expected compute the marginal implicit prices of urban park acres for dif- sign. The median household income of the purchaser was positively ferent submarkets, while the second stage demand function was and significantly related to demand. Although, the demand was estimated using the pool dataset. inelastic on income ( = 0.43), it is still the most important pre- The results from the Two Stage Least Square estimation of dictor of demand for park acres after the price of the park size, demand function are presented in Table 5. Most variables were sig- and price of living area. This is understandable given the fact that nificant and possessed the expected sign. It should be noted that previous studies (Boyle et al., 1999) concluded and our analysis con- the standard errors were robust (i.e., corrected for heteroscedas- firmed that living area is a close substitute of amenities like urban parks. 4 We have not presented the regression output for individual submarket here because of too much detail. However, those are available from the authors upon 5 A White test rejected the null hypothesis of homoscedasticity in demand regres- request. sion (Chi-Square Statistic = 1446.41, critical value = 135.81). 982 N.C. Poudyal et al. / Land Use Policy 26 (2009) 975–983 The median age of the resident was positively and strongly negatively related to the park size. A fairly inelastic demand curve related to the demand, whereas the square of this variable had a was derived with a price elasticity of −0.84. The demand was inelas- negative effect on demand. A possible explanation of this might be tic for income as well, but income was still the most important related to the declining mobility of senior citizens to properly con- predictor after hedonic prices of park size itself and its close sub- sume those resources. Residents demand nearby parks while they stitute. The study also confirmed that the price of the living space remain physically active, but that desire might diminish with aging and proximity to the nearest park were substitutes for the acres of and physical inactivity. Counter to our expectation, the percent- nearby urban parks. The demand for urban park size increases as the age of residents with a college degree was negatively related to the cost of living space increases. This might be a useful implication in demand for the park area. That might be, however, associated with landuse planning and urban sprawl management because preserv- the higher opportunity cost of time, a greater ability to afford alter- ing public open spaces could encourage high-density development native natural amenities, and other factors that were not accounted and help discourage sprawl. Similarly, residents prefer the residen- for in our demand model. tial locations by trading the size of the urban recreation parks with Race was another predictor of demand for park acres. As the the proximity of those parks. results reveal, the percentage of African–Americans was negatively Welfare analysis in this study also suggests that increasing the related to park acres at the 10% level, suggesting that non-whites current mean size of urban forest acres by 20% in Roanoke will are less likely to demand park acres. The coefficient on the square of increase the per household consumer surplus by $160. Properties the percentage of residents who are African–American was positive located within an immediate neighborhood of these parks will have but was not statistically significant. an increased total consumer surplus of $6.5 million from this pol- Estimation of the second stage demand in hedonic framework icy in the city. This estimated welfare impact might be helpful in allows us to evaluate the welfare impacts of alternative policy inter- justifying investment on open space preservation and park man- ventions. As Boyle et al. (1999) and Brasington and Hite (2005) agement, and may provide guidance in designing citizen-financed estimated for other environmental attributes, we estimated con- open space preservation or park management for Roanoke in partic- sumer surplus (CS) per household from a policy that provides more ular and hundreds of cities nationwide with similar characteristics. urban acreages under recreation parks. Increasing the current aver- Once it is known that supplying more acres for public recreation age size of the urban recreation park (35.13 acres) in the city by increases resident welfare, and provided the urban homeowners 20% (42.15 acres) resulted in an increase in consumer surplus of are willing to pay for this amenity, urban landuse planners in $160 per household. Since the values of park-related natural ameni- conjunction with neighborhood association (Lipscomb, 2003) can ties are localized within the neighborhoods, we estimated the total establish cooperative funds to establish new parks or expand exist- consumer surplus effect of this size increase on properties within ing one. Alternative citizen finance approaches could be in the form a mile6 from the park boundaries. Total consumer surplus from of one-time matching fund, partnering local government agencies increasing the current size of parks by 20% becomes $6.5 million, with residents for outright purchase of recreational land easements. for 40,984 properties located within a mile from these parks. Lipscomb (2003) suggested the concept of a “neighborhood bank” Even though there are no other studies on urban park demand as a financial platform, using a revolving low-interest fund for to compare our results, the estimated figures do not dramatically landscaping and amenity improvement in the neighborhood. Sim- differ from other related open space studies. For instance, Bolitzer ilar cooperative bonus schemes could be promoted to fund public and Netusil (2000) estimated that each additional acre of natural recreation and ecosystem service consumption on private lands area increases the nearby house price by $28, suggesting a surplus (Goldman et al., 2007), particularly in those urban neighborhoods of $196 for an increase that is equivalent to the one we discussed where a shortage of public lands already exists. above. A contingent valuation study by Breffle et al. (1998) in Boul- We also confirmed that the second stage demand estimation der, Colorado found $302 as a household’s one time willingness to could yield theoretically consistent results, provided the submar- pay to preserve 5.5 acres of open space in the neighborhood. How- kets are identified systematically. As a result, our method offers ever, the observed difference between our results and theirs may a useful approach to estimate the demand for other environmen- be because of the estimation method employed and also the con- tal amenities for which data from multiple cities are not available. text and level of urbanization between the study areas. In addition, Given that the federal and local governments are attempting to the observed differences could also be attributed to the possible preserve more open spaces in urbanizing communities, this study difference in service provided by different types of open spaces. can assist in understanding how residents respond to different lev- els of open space; and to ensure that any proposed investments Conclusion in new acquisitions can be justified by the anticipated welfare gains. As American cities continue to grow, an increasing number of residents are placing unprecedented demands on the existing open References spaces and urban recreation parks. This means urban parks will become more congested, possibly to a point which exceeds the Acharya, G., Bennett, L.L., 2001. Valuing open space and land-use patterns in urban watersheds. Journal of Real Estate Finance and Economics 22, 221–237. potential of these parks to offer recreational and amenity benefits. Allen, M.T., Springer, T.M., Waller, N.G., 1995. Implicit pricing across residential rental Using a two-stage hedonic model, this study estimated the amenity submarkets. Journal of Real Estate Finance and Economics 11, 137–151. value of and demand for urban recreation park acres in Roanoke, Anderson, S.T., West, S.E., 2006. Open space, residential property values, and spatial context. Regional Science and Urban Economics 36, 773–789. Virginia. Bolitzer, B., Netusil, N.R., 2000. The impact of open spaces on property values in Findings suggest that both the proximity to and size of the urban Portland, Oregon. Journal of Environmental Management 59, 185–193. recreation parks have a small but significant positive effect on prop- Bourassa, S.C., Hamelink, F., Hoesli, M., MacGregir, B.D., 1999. Defining housing sub- markets. Journal of Housing Economics 8, 160–183. erty values. Further, the hedonic price of an urban park acre was Bourassa, S.C., Hoesli, M., Peng, V.S., 2003. Do housing submarkets really matter? Journal of Housing Economics 12, 12–28. Boyle, K.J., Poor, P.J., Taylor, L.O., 1999. Estimating the demand for protecting fresh- water lakes from eutrophication. American Journal of Agriculture Economics 81, 6 Acharya and Bennett (2001) argued that one mile radius circle represents a typ- 1118–1122. ical walkable neighborhood around the house. We used ArcGIS 9.2 to find out the Brasington, D.M., Hite, D., 2005. Demand for environmental quality: a spatial hedonic number of properties located within that distance from the park. analysis. Regional Science and Urban Economics 35, 57–82.  N.C. Poudyal et al. / Land Use Policy 26 (2009) 975–983 983 Breffle, W.S., Morey, E.R., Lodder, T.S., 1998. Using contingent valuation to estimate a Mahan, B.L., Polasky, S., Adams, R.M., 2000. Valuing urban wetlands: a property price neighborhood’s willingness to pay to preserve undeveloped land. Urban Studies approach. Land Economics 76 (1), 100–113. 35 (4), 715–727. Mansfield, C., Pattanayak, S.K., McDow, W., McDonald, R., Halpin, O., 2005. Shades Chattopadhaya, S., 1999. Estimating the demand for air quality: new evidence based of green: measuring the value of urban forests in the housing market. Journal of on the Chicago housing market. Land Economics 75 (1), 22–38. Forest Economics 11 (3), 177–199. Cho, S., Poudyal, N.C., Lambert, D., 2008. Estimating spatially varying effects of urban McConnell, V., Walls, M., 2005. The value of open space: evidence from studies of growth boundaries on land development and land value. Land Use Policy (25), non-market benefits. Resources for the Future, January 2005. Washington, DC. 320–329. McGarigal, K., Cushman, S., Stafford, S., 2000. Multivariate Statistics for Wildlife and Chen, Z., Cho, S., Poudyal, N.C., Roberts, R.K., 2009. Forecasting housing prices under Ecology Research. Springer Verlag, New York. different market segmentation assumptions. Urban Studies 46 (1), 167–187. McPherson, E.G., 2006. Urban forestry in North America. Renewable Resources Jour- Clark, D.E., Cosgrove, J.C., 1990. Hedonic prices, identification, and the demand for nal 2006 (Autumn), 8–12. public safety. Journal of Regional Science 30 (1), 105–121. Metropolitan Design Center, 2004. Designing Small Urban Parks: A Manual for Clark, J., Kruidenier, W., Wolf, J., 2007. A national research plan for urban forestry Addressing Social and Ecological Concerns. Metropolitan Design Center, Uni- 2005–2015. National Urban and Community Forestry Advisory Council. USDA versity of Minnesota, Minneapolis, MN. Forest Service. Moore, D.S., McCabe, G.P., 2003. Introduction to the Practice of Statistics, fourth ed. Day, B., 2003. Submarket identification in property markets: a hedonic housing price W.H. Freeman and Co., New York. model for Glasgow. Working Paper. The Center for Social and Economic Research Morancho, A.B., 2003. A hedonic valuation of urban green areas. Landscape and on the Global Environment, School of Environmental Science, University of East Urban Planning 66, 35–41. Anglia, UK. Nowak, D.G., McPherson, E.G., 1993. Quantifying the impact of trees: the Chicago Day, B., Bateman, I., Lake, I., 2007. Beyond implicit prices: recovering theoretically urban forest climate project. Unasylva: Journal of Urban and Peri-Urban Forestry consistent and transferable values for noise avoidance from a hedonic property 173. Online URL: http://www.fao.org/docrep/u9300e/u9300e08.htm. price model. Environmental and Resource Economics 37, 211–232. OFHEO, 2006. Office of Federal Housing Enterprise Oversight. Online URL: http:// Ekeland, I., Heckman, J.J., Nesheim, L., 2002. Identifying hedonic models. The Amer- www.ofheo.gov. ican Economics Review 92 (2), 304–309. Paez, A., Long, F., Farber, S., 2008. Moving window approaches for hedonic price Ekeland, I., Heckman, J.J., Nesheim, L., 2004. Identification and estimation of hedonic estimation: an empirical comparison of modeling techniques. Urban Studies 45 models. The Journal of Political Economy 112, 60–109. (8), 1565–1581. Freeman, A.M., 1993. The measurement of environmental and resource values, the- Palmquist, R.B., 1984. Estimating the demand for the characteristics of housing. The ory and methods. Resources for the Future, Washington DC. Review of Economics and Statistics 66 (3), 394–404. Geoghegan, J., 2002. The value of open spaces in residential land use. Land Use Policy Palmquist, R.B., 2005. Property value models. In: Maler, K., Vincent, J. (Eds.), Hand- 19, 91–98. book of Environmental Economics, vol. 2. North Holland. Goldman, R.L., Thompson, B.H., Daily, G.C., 2007. Institutional incentives for man- PLC, 2004. Partnership for Livable Communities. Online URL: www.mostlivable.org. aging the landscape: inducing cooperation for the production of ecosystem Rosen, S., 1974. Hedonic prices and implicit prices: product differentiation in pure services. Ecological Economics 64, 333–343. competition. The Journal of Political Economy 82 (1), 34–35. Goodman, A.C., Thibodeau, T.G., 1998. Housing market segmentation. Journal of Salazar, S.S., Menendez, L.G., 2007. Estimating the non-market benefits of urban Housing Economics 7, 121–143. park: does proximity matter? Land Use Policy 24, 296–305. Green, S.B., Salkind, N.J., 2003. Using SPSS for Windows and Macintosh: Analyzing Schnare, A., Struyk, R., 1976. Segmentation in urban housing markets. Journal of and Understanding Data, third ed. Prentice Hall, Upper Saddle River, New Jersey. Urban Economics 3, 146–166. Greene, W.H., 2003. Econometric Analysis, fifth ed. Pearson Education Inc. Strong, N.A., Jacobson, M.G., 2005. Assessing agro-forestry adoption potential uti- Grisby, W., Baratz, M., Galster, G., Maclennan, D., 1987. The dynamics of neighbor- lizing market segmentation: a case study in Pennsylvania. Small Scale Forest hood change and decline. Progress in Planning 28 (1), 1–76. Economics, Management and Policy 4 (2), 215–228. Hobden, D.W., Laughton, G.E., Morgan, K.E., 2004. Green space borders—a tangi- Tajima, K., 2003. New estimates of the demand for urban green space: implications ble benefit? Evidence from four neighborhoods in Surrey, British Columbia, for valuing the environmental benefits of Boston’s Big dig project. Journal of 1980–2001. Land Use Policy 21, 129–138. Urban Affairs 25 (5), 641–655. Kline, J.D., 2006. Public demand for preserving local open space. Society and Natural Taylor, L., 2003. In: Champ, P., Boyle, K., Brown, T. (Eds.), The Hedonic Method Resources 19, 645–659. in A primer on Non-Market Valuation. Kluwer Academic Publishers, Boston, Kristrom, B., Riera, P., 1996. Is the income elasticity of environmental improvements MA. less than one? Environmental and Resource Economics 7, 45–55. Thorsnes, P., 2002. The value of a suburban forest preserves: estimates from sales of Lipscomb, C.A., 2003. Resolving the aggregation problem that plagues the hedo- vacant residential building lots. Land Economics 78 (3), 426–441. nic pricing method. Ph.D. Dissertation. Georgia Institute of Technology, Atlanta, Trust for Public Land, 2005. LandVote 2004. Online URL: http://www.lta.org/ Georgia (Chapter 1). publicpolicy/landvote 2004.pdf. Lipscomb, C.A., 2006. An alternative spatial hedonic estimation approach. Journal of Tyrvainen, L., Miettinen, A., 2000. Property prices and urban forest amenities. Journal Housing Research 15 (2), 143–160. of Environmental Economics and Management 39, 205–223. Lipscomb, C.A., Farmer, M.C., 2005. Household diversity and market segmentation Tyrvainen, L., 1997. The amenity value of the urban forest: an application of the within a single neighborhood. Annals of Regional Science 39, 791–810. hedonic pricing method. Landscape and Urban Planning 37, 211–222. Lutzenhiser, M., Netusil, N.R., 2001. The effect of open spaces on a home’s sale price. US Census, 2000. US Census Bureau report. Washington, DC. Online URL: Contemporary Economics Policy 19 (3), 291–298. www.census.gov. MacDonald, D.N., White, H.L., Taube, P.M., Huth, W.L., 1990. Flood hazard pricing White, H., 1980. A heteroscedasticity-consistent covariance matrix estimator and a and insurance premium differentials: evidence from the housing market. The direct test for heteroscedasticity. Econometrica 48 (4), 847–938. Journal of Risk and Insurance 57, 654–663.