1 THE ECONOMIC IMPACT OF PARKS ON RESIDENTIAL PROPERTY VALUES : EVIDENCE FROM GAINESVILLE F LORIDA By BRITTANY ANN MCMULLEN A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQ UIREMENTS FOR THE DEGREE OF MASTER OF ARTS IN URBAN AND REGIONAL PLANNING UNIVERSITY OF FLORIDA 2011
2 2011 Brittany Ann McMullen
3 To my grandpar ents, who always supported my academic endeavors
4 ACKNOWLEDGMENTS First and fo remost, I thank my Chair, Andres Blanco, and my Co Chair, Paul Zwick I learned and accomplished more than I thought I could as a result of their guidance. I also thank my peers who supported me in this end eavor, especially Gareth Hanley who helped me b rainstorm ideas and provided encouragement when I needed it most and for always providing a listening ear and being my work buddy for countless nights at Panera. I also thank all of the teachers who have guided me throughout the durati on of my educational experience, from kindergarten to graduate school. I truly could not have accomplished so much without them. Finally, I thank all of my friends and family who encou r aged me throughout the process. I appreciate every person who helped in s ome way wi th this undertaking, no matter the magnitude of their contribution.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 8 ABSTRACT ................................ ................................ ................................ ................... 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 12 Problem ................................ ................................ ................................ .................. 12 Research Questions and Objectives ................................ ................................ ....... 17 Method ................................ ................................ ................................ .................... 18 2 LITERATURE REVIEW ................................ ................................ .......................... 19 Introduction ................................ ................................ ................................ ............. 19 Valuing Residential Property ................................ ................................ ................... 19 Hedonic Pricing Models Case Studies ................................ ................................ 28 3 METHODOLOGY ................................ ................................ ................................ ... 39 Case Study ................................ ................................ ................................ ............. 39 Data ................................ ................................ ................................ ........................ 41 Regression Analysis ................................ ................................ ............................... 43 4 FINDINGS AND RESULTS ................................ ................................ ..................... 49 Ordinary Least Squares (OLS) ................................ ................................ ............... 49 Geographically Weighted Regression (GWR) ................................ ......................... 55 5 DISCUSSIONS AND CONCLUSIONS ................................ ................................ ... 71 Discussions ................................ ................................ ................................ ............. 71 Limitations ................................ ................................ ................................ ............... 74 Conclusions ................................ ................................ ................................ ............ 75 APPENDIX A RESULTS OF ORDINARY LEAST SQUARES REGRESSION ANALYSIS ........... 78 B RESULTS OF GEOGRAPHICALLY WEIGHTED REGRESSION ANALYSIS ........ 80
6 C PARKS IN GAINESVILLE, FL ................................ ................................ ................. 85 LIST OF REFERENCES ................................ ................................ ............................... 88 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 91
7 LIST OF TABLES Table page 2 4 Price determinants after eliminating collinear re gressors ................................ .. 37 3 1 Park ca tegories in study area. ................................ ................................ ............ 43 3 2 Variables used in regression analysis. ................................ ............................... 45 3 3 Descriptive statistics of dataset. ................................ ................................ ......... 45 4 1 Descriptive statistics of properties located within 500ft of a park. ....................... 49 4 2 OLS results for properties within 500ft of a park. ................................ ................ 50 4 3 Descriptive statistics of properties located within mile of a park. .................... 50 4 4 OLS results for properties within miles of a park. ................................ ........... 51 4 5 Descriptive statistics of properties located within mile of a park. .................... 52 4 6 OLS results for properties within mile of a park. ................................ ............. 52 4 7 OLS results for properties within mile of a park (alternative model). ............... 55 4 8 Descriptive statistics for revised dataset after geographically weighted regression ( GWR ) analysis ................................ ................................ ................ 56
8 LIST OF FIGURES Figure page Figure 2 1. Determinants of property values. (Freeman 2003) ................................ .... 20 Figure 2 2. Determinants of property values (expanded). (Nicholls, 2002) ................. 21 Figure 2 3. The Positive and Negative Impacts of Parks on Residential Property Values. (Crompton, 2005) ................................ ................................ ........ 23 Figure 2 4. Implications for the proximate preminus. (Crompton, 2005, p. 210) .......... 38 Figure 3 1. Map of Study Area ................................ ................................ ..................... 40 Figure 3 2. Map of study area minority percentages by census block. ....................... 4 7 Figure 3 3. Map of study area population density by census block. ........................... 48 Figure 4 1. Park distance correlation coefficients for properties within mile search radius. ................................ ................................ ................................ ....... 59 Figure 4 2. Park distance correlation coefficients Cofrin Nature Park. ....................... 61 Figure 4 3. P ark distance correlation coefficients for properties within mile of Cofrin Nature Park. ................................ ................................ ................... 62 Figure 4 4. Graduated p ark distance correlation coefficients for properties within mile of Cofrin Nature Park. ................................ ................................ ........ 63 Figure 4 5. Park distance correlation coefficients for properties within mile of Springtree Park. ................................ ................................ ........................ 66 Figure 4 6. Graduated park distance correlation coefficients fo r properties within mile of Springtree Park. ................................ ................................ ............. 67 Figure 4 7. Park distance correlation coefficients for properties within mile of the Boys Club of Alachua County. ................................ ................................ ... 69 Figure 4 8. Graduate park distance correlation coefficients for properties within mile of the Boys Club of Alachua County. ................................ ................. 70 Figure A 1. Ordinary Least Squares results, 500ft search radius ................................ .. 78 Figure A 2. Ordinary Least Squares results, mile search radius. .............................. 78 Figure A 3 Ordinary Least Squares results, mile search radius ................................ 79
9 Figure A 4. Ordinary least squares results, using alternative explanatory variables, mile search radius. ................................ ................................ ................ 79 Figure B 1. Geographically Weighted Regression Results for properties within a mile search radius ................................ ................................ .................... 80 Figure B 2. Geographically Weighted Regression Results Frequency Distributions. ................................ ................................ ............................. 83 Figure B 3. Geographically Weighted Regression Results mile search radius for alternative model including additional explanatory variables. .................... 84
10 Abstract of Thesis Presented to the G raduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Arts in Urban and Regional Planning THE IMPACT OF PARKS ON RESIDENTIAL PROPERTY VALUES: EVIDENCE FROM GAINESVILLE, FL ORIDA By Brittany Ann McMullen December 2011 Chair: Andres Blanco Co chair : Paul Zwick Major: Urban and Regional Planning Human behavior will always have an im pact on the natural environment H owever, actions taken by local go vernments can work t o reduce the imp act and provide for sustainable communities. Through the creation and maintenance of parks, natural resources are preserved, and result in the provision of ecosystem services for communities. In addition to serving environmental sustainability interests parks contribute to our overall phys ical and psychology well being. Factors such as air quality, water quality, access to recreation al opportunities and existence of spaces for community engagement affect o ur physical and mental health. Despite the ben efits that parks provide, b ud get constraints can present a challenge to providing these necessities Because of this, there is a need to quantify the benefits that parks provide Capitalization of park value into housing prices is one way to accomplis h t his T he goal of this study was to estimate the economic impact of parks on residential property values using Gainesville, FL as a case study Relying on principles established in reviewed literature, regression analysis was utilized to me asure the corr elation between market value of single family res idential homes and proxi mity to the nearest park.
11 Results showed that in some instances proximity to a park d oes increase the value of a home; however, t he impact varies by location. Further consideratio n of the use of the park is needed to better understand the impact. These findings indicate that parks used primarily for recreation do not have a positive impact on property values; negative externalities such as noise or traffic make it less desirable t o live in close pro ximity to these facilities. Conversely, living next to a natural resource or leisure activity based park does have a po sitive impact on housing value. Overall, there is potential for citizens and local governments to receive economic b enefits a s a result of the preservation of open spaces in communities. Home owners benefit from higher property va lues, and local governments receive an increased amount of property tax revenue as a result.
12 CHAPTER 1 INTRODUCTION Problem Recent acknowle dgement of environmental issues such as climate change and resource depletion has prompted cities to contemplate how they can address sustainability in the built environment The consideration of environmental issues can been guided by four laws of ecolog y, established by Barry Commoner in his 1971 book The Closing Circle. These laws remain timeless : everything is connected to everything else, everything must go somewhere, nature knows best, and there is no such thing as a free lunch. These basic princip les have been used as a guiding poin t for many environmental issues and refer to entity outside and separate from people, but something th at we are a part of and exist symbiotically with The idea s that everything is connecte d to everything else seem to have dissipated from the minds of some policy makers and others who influence the design and maintenance of the built environment Although interest in e nvironmental issues has recently increased in p ast years strategies for sustainability and regeneration i n urban settings have chiefly focus ed on man made and built components (Chiesura, 2003 p.129 ). Permanently preserved open spaces such as parks can be used as a mechanism to address the need for env i ronmentally sustainable cities. At the microenvironment level, t hey provide ecosystem services such as stor mwater control and air control. T he existence of soil and natural resources in areas that may otherwise be filled in with concrete or asphalt aide s in the natural regulati on of rainfall, which affects water quality and runoff issues In addition, trees act as n atural air quality control ( Costanza, 1997) Open spaces are also essential in temperature
13 regulation; trees and plants provide shade and m aintain a certain degree of humidity (Morancho, 2003). They also provide for the preservation of habitats, especially when design is focused on the connectivity of existing and future parks. Aside from the numerous benefits for habitats and natural reso urces, permanently preserved open spaces, such as parks, play a n important role in addressing physical and psychological human health issues in urban settings. In regards to physical health, t he recreational ben efits that parks provide for communities are obvious. They promote healthy living by providing for the opportunity to exercise and engage in physical activities The existence of parks, playgrounds, and organized sports in neighborhoods has been show n to play a key role in the pr evention of childho od and adolescent obesity ( Goodman, 2002) an issue that is gaining increased attention across the country Not only do parks in urban settings impact physical health through the provision of recreational opportunities, they are also a contributing factor to psychological wellbeing ; i t has been argued that simply coming into contact with nature is a basi c human need (Maller, 2003). E vidence has shown that the existence of natural resources in cities has the potential to reduce stress and foster ov erall ps ychological well being by prov from urban living (Parsons, 1991 ). Some research has gone so far as to sug gest that having access to natural resources is not only important, but essential for long term human health and development ( Driver e t al. 1996 ) In addition, evidence has shown that the existence of natural settings in urban locations can reduce lead to lower levels of fear in residents, and behavior that is less violent. P arks also provide an outlet for community engagement through recreational activities and
14 opportunities for interacting in informal meeting places As Baum (1999) states, a key component of healthy communiti es is the provision of various opportunities for citi zens to interact both formally and informally with each other; parks serve this interest. overall than other individuals, and the long (Maller, 2009, p. 66). In short, permanent ly preserved open spaces in cities address environmental sustainability and human health issu es in ways that man made components cannot. The idea that parks contribute positively to health is not a new one. Physical and psychological health justifications used today were used when parks were first designed in the ninet eenth century. Many parks were developed b ased on the hope that they cou ld provide positive physical and psychological health advantages, such as reduction of disease, recreational opportunities, and reduction of stresses associated with urban (Maller et al., 2009 p. 54) Although the importance of parks in urban settings h as been a notion recognized since the early stages of urban planning in the United States, th e provision and maintenance of these resources i s not always a top priority w hen other more pre ssing issues are present and budgets are limited However as Maller states Because our water quality, air quality, economic vitality, and personal well being are as dependent on natural resources as they are on transportation, communicatio ns, and public safety systems, parks, by providing access to nature and protecting ecosystems, are an essential part of the infrastructure of our cities and communities. (p. 55) Based on evidence of the importance of natural settings in urban environments there are important questions to be answered about how to make environmental efforts
15 appealing to governments businesses, and individuals. While some make efforts to be environmentally responsible simply because they are aw are of the necessity to do so many do n ot s ee enough incentive to employ environmentally friendly practices or may even find it cost prohibitive. F actors such as l imited budgets can lead to a deficit in the existence and maintenance of parks and open spaces in a community ( Tyrvinen and Vnnen 1998). Aside from the fact that parks are a public amenity that are not a pressing issue compared to things such as safety or transportation, they are often seen as an investment that offers little to no financial return. Therefore, it is imperative that financial justification for parks and open spaces be provided in order for communities to find reason to succ essfully support these resources This leads to the major impetus for this research: how can the value of things such as improved water or air quality, protection of a habitat, flood control, the opportunity to interact with nature, or the various other physical and psychological health benefits to a community be quantified? Placing an estimated market value on a nonmarket good is diffi cult While studies attempting to put an objective monetary value on parks have emerged largely in the last 45 years or so, the concept of developing parks with at least the partial intention of reaping economic benefits is in fact not a new concept. The initial development of parks in America was partially driven by the expectation of economic contributions to city tax revenues (Nicholls, and Crompton, 2005). However, this concept seems to have become less recognized in recent years, with parks bei ng held low on the priority list during times of budget shortfalls. Much of the earlier research on this topic arose because of the fact that, although there was a general consensus that parks had a positive influence on the real estate values of the
16 prop erties around them, there was a lack of evidence to back up the theory. Therefore, there was a desire to develop an objective method of evaluation in order to more accurately account for the benefits provided by such investments and provide this informati on to the public. Calculating the market value o f the human health and ecosystem services offered by parks and open spaces can be approached using a number of methods. There are two commonly used approaches. The first is the use of consumer surveys, wh ich ask consumers to rank the importance of park or open space location in their decisio n to move to or visit an area. A 1995 study conducted by American Lives, Inc. found that a significant number or home buyers in a study area (77.7 percent) ranked open Dhanju, 2006 ). Simply considering the fact that properties for sale are often advertised as having the benefit of nearby natural amenities is an indicator of the desirableness of these resource s to consumers (Crompton, 2005). The second method used to estimate value is based on statistical theories, and relies on the use o f hedonic pricing models. This method attempts to capitalize the value of nearby open spaces and other variab les into property price In a competitive housing market, prospective homebuyers will bid up the prices of homes with desirable surroundings, thereby capitalizing such externalities into the price (Anderson and West, 2006). Many s tudies utilizing hedonic methods have shown a positive relationship between recreational features and increased valuations of properties located near them. Crompton (2005) suggests that houses located in close proximity to a park can receive an added value of up to 20% (p. 203).
17 In addition to benefiting home owners, the increase in property values in turn results in an increase in government revenues due to the subsequent increase in property taxes. Some literature suggests that the increase in property tax revenues can absorb the cost of designating and maintaining the park itself (Crompton, 2000 ) Furthermore, some evidence proposes that the preservation of open space can be an alternative that is less expensive than development of the land because of the infrastructure that goes along with building homes (schools, roads, utilities, etc ) In this sense, it can be argued that parks and open spaces are essential in maintaining economic stability of communities rather than a drain on the budget (Crompton, 2000 ). It is for the se reasons, Crompton argues that creating and maintaining parks in communities can be economically sound Research Questions and Objectives The purpose of this research was to examine whether or not parks have a n economic impact on nearby residential p roperties, and what potential economic benefits can be seen by the community as a result. R esearch is based on the following questions: Do parks have an economic impact on proximate residential property values? If so, is the impact positive or negative? Through the use of qualitative and quantitative analysis, usi ng Gainesville FL as a case study, th ese q uestions are answered. T he hypothesis is that there is a negative correlation between distance to parks and property values; as dist ance goes up price is expected to go down. The relationship is expected to be nonlinear because of the positive and negative externalities resulting from the park
18 Method Housing is considered a multi attribute good H edonic pricing models or regression a na lyses are us ed to estimate the economic value that each attribute contributes to the total price of the property. Analysis for this study was centered on the estimation of the value of a park located within a specified distance of residential properties measuring di stance to the nearest park in feet Two regression analysis methods were utilized : ordinary least squares (OLS) and geographically weigh ted regression (GWR) Beginning with the ordinary least squares regression method, the relationship b etween the dependent variabl e ( market value ) and independent variables relating to structural, neighborhood and park location attributes was tested. Using independent variables that were discovered to be statistically significant g eographically weighted regres sion analysis was th en performed to improve upon the results. The use of this method allowed for a more in depth analysis of how specific parks of different types impact property values at different locations.
19 CHAPTER 2 LITERATURE REVIEW Introduction This chapter contains information substantiating t he background and methods of this study. The first portion provides a basis for understanding t he variables that affect residential property values. This includes a review of the basic components of property price, a s well as a discussion of how a specific variable such as distance to a park impacts price. After establishing core principals, a review of cas e studies which have utilized hedonic pricing models is provided to allow for a better understanding of the ap plication of these models, and to provide support for the hypothesis that parks have a positive economic impact on proximate property values. Valuing Residential Property To estimate the impact that an independent variable such as distance to a park has on t he value of residential property it is important to first have an understanding of the numerous variables that affect housing values Housing is a multi attribute good (Morancho, 2003 ) I n other words, arriving at a final value for housing price takes into account a b undle of attributes. While the specific variables that are used may differ from one study to the next there are general charac teristics that apply to them all According to Freeman ( 2003 ), h ousing prices are based on three basic categor ies of attributes : structural, neighborhood, and environmental (see figure 2 1)
20 Figure 2 1. Determinants of property values (Freeman 2003) In this model, s tructural characteristics include features of the housing unit s uch as age of the house, numbe r of rooms, and type of construction. Neighborhood characteristics include features of the area in which the house i s located. This can include quality of neighborhood schools, crime rates in the area, and accessibility to amenities such as parks, stores and work places. Finally in this model environmental characteristics are location sp ecific environmenta l amenities, such as air quality, water quality, etc. As Freeman notes, there is a challenge with assigning value to some environmental characterist ics. For example, it is difficult to assign an exact number to a factor such as air quality as it changes over time due to things such as emission levels or meteorological factors (Freeman, 2003). This speaks to the problem addressed by this research. Nicholls (2002) expands on this concept, defining property price to be equal to structural, neighborhood, community, time related, environmental, and locational attributes (see figure 2 2)
21 Figure 2 2. Determinants of property values (expanded) (Nich olls, 2002) His definition of structural attributes is similar t o the definitio n provided by Freeman with a few additional considerations In Nicholls definition, structural attributes include number of bedrooms, number of bathrooms, square footage, exi stence of air conditioning/heat and parcel size. Neighborhood attributes differ model however, account ing for variables such as the socio economic and demographic characteristics of an area. For ex ample, average income poverty levels, et hnic composition, age, etc. would fall into this category. Neighborhood attributes can also include characteristics such as quality of neighboring structures, ownership versus rental compo sition, and population density. The third variable discussed by Ni cholls is c ommuni ty attributes. These are factors such as school and tax districts. These do not need to be considered in areas which share the same attributes in this regard. T ime related attributes are those which affect pr operty values as they relate to the real estate market as a whole. These include the month and day of a sale and the length of time that t he property was on the market. Environmental attributes are along the same lines
22 as t hose discussed by Freeman and include factors such as the view from a property, nois e levels, pollution levels, and stormwater control. Locational attributes are the final varia ble discussed by Nicholls and include characteristics that Freeman inclu des in the neighborhood characteristics category These includ e things such as proximity and acc essibility to various amenities or services. Brigham (1965) terms this the ameni ty value As he states, the level of an amenity value is undoubtedly a factor that is subjectively determined by every individual Proximit y to such amenities can have a positive or negative effect on property values For example properties located in close proximity to a waste site would be expected to be negative ly a ffect ed by such an amenity while properties located near public transpor tation would be expected to exhibit positive benefits Other e xamples of locational attributes provided by Nicholls include access to p ower lines, highways, shopping centers, and employment opportunities Estimating the impact of parks in particular can be accomplished by using characteristics of parks, such as distance, for one of the explanatory variables. The idea that parks have a positive impact on property values can be referred to as the proximate principle whic h, as Crompton (2005) describes, i s based on the th eory that people are often willing to pay more for a home located close to a park than they are for a similar home that is located farther away previously cited formula, parks fall into the category of either location al attributes or neighborhood attributes, in the context of accessibility to amenities. Parks could also be considered in environmental attributes as well because of their provision of views of nature and improved quality of things such as pollution level s and stormwater control. In the cases
23 in which distance has been shown to increase values of proximate properties, Crompton suggests that the impact is not equally seen by all properties surrounding a park; this h e The pos itive impact to properties located directly adjacent to a park may be lower than the impact on properties located a sho rt distance away (see figure 2 3) Homes located on the edge of a park may be subject to nuisances created by access points or the noise th at results from park activities, thereby affecting the degree to which the properties benefit from the positive externalities. Figure 2 3. The Positive and Negative Impacts of Parks on Residential Property Values. (Crompton, 2005)
24 As previousl y discussed, the price of a good can be considered as the sum of the price paid for each individual characteristic The use of regression models, m ore specifically hedonic pricing models, allows for the estimation of the marginal implicit price of a speci fic characteristic related to a differentiated market good (Irwin, 2002, p.466) which represents the marginal willingness to pay for a specific characteristic (Crompton, 2005). Put differently, these models allow one to isolate and quantify the effect that a specific variable or variables have on the price of a market good. In terms of real estate valuation t hese models assume that ho using prices are in equilibrium, meaning that individual consumers have purchased a home which is the most utili ty maximizing for them, given the prices of other housing locations, and that the price just clear s the market considering the characteristics of the house and the existing stock of hous ing in the area (Freeman, 2003, p.357). In addition, these models tra ditionally assume that the study area is representative of a single market for housing (Bol itzer and Netusil, 2000, p.186). Freeman (2003) provides the following representation of a hedonic pricing model, based on theories proposed by Rubinfeld (1977), Po linsky and Rubinfeld (1977) and Witte et al. (1979): Rh j = Rh(S j N j Q j ) (2 1) Where: ( Rh j ) = Price of the j th residential location ( S ) = Structural characteristics ( N ) = Neighborhood characteris tics ( Q) = Environmental characteristics. Nich olls (2002) uses a linear form of the model derived from the same premise: 1 s X s N X N C X C + L X L + E X E + R X R + (2 2) where:
25 P = observed property prices X s = structural attributes X N = neighborhood attributes X C = community attributes X L = locatio nal attributes X E = environmental attributes X j = time attribu tes = stochastic disturbance term (variables that are not included in the model) 1 = constant term x = estimates of relevant attributes' implicit marginal prices after differentiation. In the case of these equations, residential property val ue is the dependent variable, and the attributes that make up that value are the explanatory or independent, variables (Knaap, 1998). The hedonic variable is that which is being tested and holding all other variables constant, allows for the estimation of its in fluence on market values (Morancho, 2003, p.36). In the case of estimating the value of proximity to parks, distance to a park is the hedonic variable. T he specific variables that are used may differ depending on what is being tested and how groups of at tributes are defined but t he basic concept remains the same for these types of market price valuations Different functional forms of a hedonic pricing model can be applied. Linear forms are commonly used. As Morancho (2003) describes, i n linear method s, the willingness to pay remains constant; in other words, there is no dependence on the starting level of the marginal will ingness to pay ( p. 37). A commonly used linear modeling technique is the ordinary least squares regression method. A s explained b y Craven (2011), this method used to model a single response variable which has been recorded on at least an or to several explanatory variables as well as c ategor ical explanatory variables (p.224). It operates on the basic assumption that the relationship between the response (dependent, or Y) variable and the continuous
26 (independent, or X) variable can be represented using a line of best fit, where the resp onse variable is predicted by the continuous variable (p.224). Craven provides this equation for representing the relationship mathematically: Y x (2 3) Where : = the value of Y when X is equal to zero = the slope of the line or correlation coefficient The correlation coefficient describes the change in Y that is associated with a unit change in X These types of models present limitations. Morancho (2003) suggests that one limitation stems from the assumption that the relations hip between price and an environmental variable may be non linear. It is because of this that logarithmic forms are often used. However, linear models are often chosen b ecause the parameters are easier to interpret. A not h er common problem that is associ ated with regression models in general is multicollinearity As explained by Sundberg (2002), multicollinearity refers to exact linear relationships within a set of variables, usually explanatory. This results in unstable coefficients, which are sensitiv e to minor changes in input data. It may also result in coefficients which are not uniquely defined or have inflated variances, leading to the impossibility to interpret them independently. As Sundberg goes on to explain, eliminating collinearity problem s can be approached by eliminating one of the variables in a collinear relationship. The challenge resides in the ability to identify which variables are exhibiting multicollinearity. The box cox flexible functional form along with utilizing a goodness o f fit test has been used by many researchers to address this issue ( Anderson, 1982, p. 334 ). However, Anderson warns of four major limitations of usin g
27 the box cox functional form. The first limitation is that best fit criterion cannot be relied upon to achieve accurate estimates of characteristic prices; in fact, accuracy is reduced by the large number of coefficients estimated with the Box Cox functional form, possibly leading to inaccurate estimates of prices. Second, it is not appropriate to use with any dataset containing negative numbers because of the impossibility of ra ising a non integer number to a non integer power Third, because the result of the mean predicted value of the dependent variable may not equal the predicted value, the predicted value may be biased. Finally, the results of nonlinear transformations are often complex estimates of slopes and elasticities which are simp ly too cumbersome to use. O ne of the more problematic limitations relating to linear models and techniques such as ordinary least squares is the issue of spatial heterogeneity Hedonic pricing models assume that the study area is representative of a single market for housing and therefore impose a constant price structure on all housing units However, as Bitter (2007) states, increasing evidence has shown that more consideration should be given to spatial issues because the marginal prices of m any attributes vary over space. If spatial heterogeneity exists, stationary coefficient models will produce parameters that are in essence an all locations. A failure to incorporate spatial heterogeneity will result in biased coefficients and a loss of explanatory power and may obscure important dynamics relating to the operation of housing markets. (p. 8) To account for this issue, the geographically weighted regres sion method can be used. T his technique calculates correlation coefficients at each individual location by performing a regres sion at every observation point rat her than assigning a single coefficient that is in essence an average value of all locations. This method is a relatively new technique, having been introduced to geography literature in 1996 by
28 Brunsdon et al. (Wheeler and Pez 2010). Overall, m any opt ions exist for employing regression techniques; a s Anderson (1982) states, there is no globally accepted functional form to use in reg ards to hedonic pricing models. T herefore, he suggests the reasonable action is to try more than one form and ultima tely use the model th at shows the best performance. Hedonic Pricing Models Case Studies Many case studies attempting to estimate the added value of parks use variations of the previous ly discussed models and have shown that parks and other open spaces do hav e a positive impact on residential property values ( Bolitzer and Netusil, 2000 ; Nicholls and Crompton, 2005; Anderson and West, 2006 ). Studies in this area have explored the effect of different types of open spaces, including parks, greenways, agricultura l land, forest land, and lands with conservation easements. Their results suggest the economic impact that is realize d can be partially dependent upon t h e type of park or open space. Geoghegan (2002) suggests two basic categories of open spa ce: developab le and permanent. He p erformed a case study regression analysis to distinguish between the impacts of the two categories. By his definition, d evelopable open space includes designations such as agricultural cropland, pasture, and forest ; p ermanent open s pace includes parks and lands with conservation easements ( lands that have no chance of being developed in the future ) The results from his model showed that individuals are willing to pay more up to three times as much, to live in closer proximity to p ermanent open space than developable open space. Not only are there different categories of open space to consider, reviewed literature suggests that different categories of parks should be considered as well.
2 9 Parks are often distinguished by their size the area that they provide service to (suc h as neighborhood or community), o r use type. The following is a review of t hree case studies which utilize hedonic pricing analysis to esti mate the economic value of open spaces, mainly parks or similar open spa ce categories. The results of these studies provide evidence which supports the hypothesis that proximity to parks does have a positive economic impact on residential property values. One of t he earlier studies investigating this topic focused on a neighb orhood park in Lubbock, Texas. Kitchen and Hendon (1967) conduct ed a case study for the purpose of estimating the economic benefits derived from the location of public recreation Th ey categorized the types of benefits received from parks as primary and secondary. Primary benefits are define d as those that benefit a user of the amenity. For example, this would include the benefits that one gets from utilizing the recreational facilities of a park, or someone who enjoys the scenic view. Secondary benefi ts are those affecting the local economy such as increased tax revenues resulting from recreational developments. The focus of their study was secondary benefits The hypothesis was that the value of land diminishes the further away a property is from a park Their method involved estimating the distance to the park and performing correlations of house price and distance hous e assessed value and distance and property value and distance For thi s, they used a linear correlation analysis. As a basis f or their analysis, they selected a 10 acre neighborhood park in a residential area that was considered to be mostly homogenous. T he neighborhood included houses that were price d from $12,000 to $18,000 ( in 1967 dollars) and were relatively new at the tim e having been built since 1950. According to their observation, there were no other significant
30 amenities contributing to value in the neighborhood besides the park. They created a which was a 2 block area around the park. They e liminated an area around a neighborhood school, as well as commercial areas since those factors w ere predicted to have an impact on property values as well The zone of influence resulted in an initial 550 properties to be included in the study They co llec ted three sets of data to analyze : access distance (in feet) sale price (of pr operties sold within the prior 5 years), and assessed value, which was retrieved from the county P roperty Appraiser. The final sample set included only data regar ding prope rties that had been sold in the five years prior to the study (480 in total ) They first performed a linear analysis between access distance and value of parcels. This did not result in a significant statistical relationship between variables. Therefore, t here was no support for the hypothesis in this case. A second analysis was performed using distance and sale price of parcels. This resulted in a statistically significant relationship with a positive correlation coefficient of .0541, indicating that t he value of properties went up th e fa rther a property was from the park; opposite of the hypothesis. They offered several rea sons to explain why the two models did no t result in strong relationships First, they cited the lack of a method to test for hom ogeneity in the market area being examined. They also postulated that variations in time of development could result in weak relationships ( p.360). To address these issues, they performed a final correlation between land values alone and distance to the pa rk as opposed to looking at housing value as a whole. After running a significance test, it was concluded that the resulting coefficient of .17 was statistically significant; there was a
31 negative although small, correlation between land v alue and dis tance to the par k. In other words, land value decreased by $17 for every foot i ncrease in distance. Overall, in terms of proving their hypothesis, they concluded that land value was the only dependent variable that was impacted positively by proximity to the park. One explanation provided for this outcome was that land values are arguably more homogenous and objective than housing values which are influenced by a number of different variables and persona l preferences. As they point out, although the inf luence of park distance is relatively low in comparison to other variables influencing land values, in the consideration of the costs and benefits of parks it is significant to note this time period, provided a basis for subsequent research seeking to determine the economic impact of open spaces on property values. Studies conducted since then have yielded similar results and have provi ded evidence based on not only l and values, bu t also housing values including structural and neighborhood attributes. One such study conducted by Bolitzer and Netusil (2000) examine d the degree to which open space proximity had an eff ect on the sale price of a home in Portland, Oregon. They divided the study area into five quadrants: north, south east, north east, south west, north west. The following hedonic pricing model was used to calculate the value of open space in this study : P i = P( S i ,Q i G i O N i ) (2 4) Where: P i = Price of home S i = structural characteristics Q i = environmental characteristics N i = neighborhood characteristics.
32 As they considere d open space characteristics to be a subcategory of neighborhood characteristics, t hey subdivide d N i into a vector of open space characteris tics (G i ) and a vector of other neighborhood characteristics (On i ) Structural characteristics included age of the house, number of bathrooms number of fireplaces lot acreage and square footage of the structure A nuisance dummy variable for traffic w as included, re flecting street traffic near a house A location variable and size of the closest open space were also used To account for neighborhood characteristics, they combined the distance to the central business district with the quadrant in whic h the house was located. For their analysis, they utilized two sources of data The first was county tax assessor data of home sales that occurred between 19 90 a nd 1992. This included t he date of sale, as well as structural and neighborhood characteri stics. The resulting dataset contained 17,953 homes; however, homes with a sale price that did not indicate an arms length transaction ( sales of $1), and homes that sold for less than their assessed land value were not used. In addition, they did not incl ude records which clearly indicated recording errors. The final data set contained 16,402 samples, with a mean sale price (in 1990 dollars) of $66,000. The second set of data included information such as distance to open spaces an d the central business d istrict and was obtained from the Geographic Information System (GIS) of the elected regional government, Metro. U sing a search radius of 1500 feet (a little over a quarter mile) fro m the sample properties 218 open spaces were selected The decision to use 1500 feet stemmed from the authors definition of block size, 200 feet, multiplied by a search radius of 7.5 blocks If a home fell within 1500 feet of more than one park, the closest park was used. Open space was divided into four
33 categories: public park, private p ark, cemetery, and golf course. The category with the highest frequency was public parks, with a count of 193. The rest of the study area co nsisted of two private parks, fifteen cemeteries, and eight golf courses. Based on the assumption that proximity to open space influences marginal implicit price nonlinearly because of resulting positive and negative e xternalities they used seven zones to categorize distance: 100 feet or less, 101 400 feet, 401 700 feet, 701 1000 feet, 1001 1300 feet 1301 1500 feet and more than 1500 feet The expectation was that value would increase up to a certain point, and then would begin to decline. They teste d for three different scenarios using linear functional and semi log functional form s of the pricin g model for each. The first scenario they tested was m odel A : the effect of any type of open spac e within 1500 feet of a property using a dummy variable for the existence of open space within the specified search distance The correlation coefficient for the dummy variable was positive, and was stati stically significant in both the linear and semi log models. The linear model showed that the selling price of a home located wi thin 1500 feet of open space sold for $2,105 (in 1990 dollars) more than a home located at a distance of more than 1500 feet Open space size was also a factor, adding an additional $28.33 for every additional acre. Combining these two factors, for a home located within 1500 feet of an open space that is 20 acres, the value increase d by $2,607. The results of the semi log model showed that homes within 1500 feet of an open space would sell for 1.43% more than a home located outside of the 1500 feet zone. Size had an influence in the semi log model as well, adding $1,247 to the pric e of a home within 1500 feet of a 20 acre park.
34 Model B refined the analysis by distinguishing between the four categories of open space The linear and semi log model s re sulted in only the public park and public golf course coefficients being statically significant. The estimated added value to a home within 1500 fe et of a public park was $2,262; homes within 1500 feet of a golf course had an added value of $3,400. Private parks and cemeteries did not have a statistically significant effect. The semi lo g model showed that proximity to a public park provided an increased value of $845, and proximity to a public golf course added $3,940. Private parks and cemeteries were again not found to be s tatistically s ignificant The si ze of open space was again st atistically significant in both the linear and semi log models. U sin g mean sizes homes located near a public park had a n increase d value of $2,780 in the linear model, and a $1,360 increase in the semi log model. Using the mea n sizes of public golf cour ses, an increase of $6,408 was observed in the linear model and $6,926 in the semi log model. Model C focused on the effect of distance from an open space using the six di stance zone categories The smallest distance variable of 100 feet was included to c apture homes in very close proximity to a park; they hypothesized that the homes in this zone would be im pacted by the negative externality of noise. This model did not include traffic and size of open space variables ; reasoning for removing the traffic v ariable was based on the assumption that traffic variables in the first two models would pick up on that negative externality Open space size was not included as the sole purpose of this model was the effect of proximity to open space. Results of t his model were as expected; value increased to a certain point, and then declined the greater the
35 distance got The zone of less than 100 feet was the only zone found to be not statistically significant neither proving nor disproving their hypothesis at this distance. An important po int that is noted by Bolitzer and Netusil is t he difficulty of measuring how the positive s and negatives of marginal implicit price of open space characteristics such as proximity and type, are accounted for Residents living ne xt to a public park may experience positive externalities from their close location to the park, but may also experience the negative externalities of traffic and noise. A resident located at a somewhat farther distance, perhaps two or three blocks, may e xperience the same positive externalities due to their proximity, but not the same negative externalities as those located directly adjacent to the park These results provide evidence to support the aforementioned net effects curve Their overall conclu sion was that distance to an open space can have a statistically significant effect on the price of a home; however the level of influence i s depend ent on the type of open space In addition, t he effect that is seen can vary depending on which functional form is used (linear or semi log). Based on their results, t hey conclude d that the semi log form is more appropriate for these types of relationships ; a linear model is restricted because the relationship between the explanatory and dependent variable is not constant Morancho (2003) conducted a similar study which used a hedonic pricing model to analyze the relationship between housing prices and urban green areas. The sample consisted of 810 properties in Castelln, Spain. Explanatory v ariables influe ncing housing price include d : age number of bedrooms and bathrooms, size, existence of an elevator, number of car spaces in a garage, what floor the dwe lling was located on, whether it was an individual house versu s a studio or apartment, square meters of the
36 balcony, whether or not it was classified as protected by public policies ( thereby making it more fiscally advantageous) existence of a storag e room and dista nce from the town center (in meters). T hree separate hedonic variables were considered in regards to parks : a dummy variable for view of a garden or public park (assigning a value of 1 if the house overlooks a garden or public park) size of the closest green area (in m eters squared ) and distance to the nearest green space (in m eters ) The fo llowing model was used as a basis for analysis, and applies the same theory as the previously discussed models : P = f(x 1 x 2 x n z ) (2 5) Assuming a linear model, the following formula was used : P i = b 1 x 1 i + b 2 x 2 i + b 3 x 3 i + + b n x n i + b z z i i i = 1, 2 ., T where x 1 i x 2 i ., x ni z i are variables describing the attributes of housing parameters b 1 b 2 ., b n b z are the marginal willingness to pay for each attribute and is the error term. The marginal willingn ess to pay for an additional unit of the environmental good z is b z They uti lized the ordinary least square s (OLS) method to analyze the data. They chose this this procedure because among hedonic models parametric specifications, OLS estimation s pr ovide similar results on the goodness of fit, the mean quadratic error and the forecasting capacity of the model to those provided by other specifications which are more difficult to interpret such as those devised by Box Cox and Wooldridge ( p. 38)
37 Ta ble 2 4 Price determinants after eliminating collinear regressors (Morancho, 2003, p. 38). Variable Coefficient t ratio SIZE 101.33 20.6 M2BAL 133.88 6.8 AGE 46.19 3.3 BATHS 2912.68 8.3 STORE 951.03 3.3 ELEVATOR 2464.51 5.5 PROTECTED 2234.60 8. 4 HOUSE 3238.75 5.2 GREENDIS 3.28 6.0 C 2554.30 3.9 R 2 = 0.743, adjusted R squared = 0.740, F ratio = 257.09, n = 810. Table 2 4 shows the results of the ana lysis using the OLS method after eliminating col l inear or insignificant variables (rooms existence of garage, unit floor, distance from town center elevator, green space views, and size of green space) V ariables found to be collinear were not elaborated upon. All coefficients were found to be statistically significant. The value for R s quared sho wed that 74 % of the price variance could be attributed to the explanatory variables. The t statistic s showed that the variable with the greatest explanatory power was housing size ; the coefficient of 101.33 meant that a larger size equaled an in creased housing price The only variable that was found to be re levant in regards to green space was distance The coefficient of 3.28 suggests that for every 100 meters, housing price fell by 328,400 pesetas. Green space size and the existence of a vi ew of green space or public park were not statistically significant. Morancho postulates that the fact that size was not significant but location in proximity to a green space was suggests that the existence of numerous small parks throughout a city may be more appropriate than a small number of larger parks. Crompton (2005) supports this idea, suggesting that the size and number of parks are important to take into account for the aggregate increase in property value
38 and subsequent increase in property taxes. Based on the results of a case study performed in the Dallas Fort Worth area of 14 neighborhood parks by Miller ( 2001), he posits that although large parks add more value to a property than a small park, the added value is small in comparison to the premium added by proximity. Therefore, a greater number of small parks will lead to a higher added value overall than one large park that does not allow for as many houses in close proximity. Figure 2 4 il lustrates this point by showing the potential for dividing 50 acres of parkland amongst 6 smaller parks rather than one large park. Figure 2 4 Implications for the proximate preminus (Crompton, 2005, p. 210)
39 CHAPTER 3 M ETHODOLOGY This study seeks to assess the economic impact of parks o n re sidential property values. Based on noted evidence, the hypothesis is that proximity to a park will have a positive economic i mpact on the value of nearby properties. The relationship is expected to be no nlinear because of the positive and negative exter nalities resulting from the proximate park. As previously mentioned there are two commonl y used methods to approach this question : household surveys and hedonic pricing or regression, models Studies using household surveys are designed to elicit perso nal values or preferences and have qualities that might be more useful in a qualitative discussion of this issue. Since the aim of this study was to quantify the value of parks and due to the fact that the reviewed literature utilized quantitative method s, regression methods were used for this analysis Ordinary least squares (OLS) regression was first utilized to t est the degree to which park distance is significant to housing prices Because this method lacks the ability to account for spatial varia nts, geo gra phically weighted regression (GWR) was subsequently used to achieve the most accurately defined model. In order to test for the impact of park distance on housing values, a case study of properties in Gainesville, FL was used. Case Study Gainesv ille is the largest city and county seat of Alachua County It is approximately 62 square miles. According to 2010 Census data, the population is 134 297 with 55 296 households, and 62 322 housing units. Population density is 2,028.5 per square mile. The median age is 26 and the average household size is 1.6. It is the location and oldest university as
40 well as Santa Fe College. Being the location of two educational institutions there is a sizeable student pop ulation: 50,116 at UF (Office of Institutional Planning and Research, 2011) and 17,630 at Santa Fe College (Santa Fe College, 2011). Figure 3 1 Map of Study Area
41 Data Two datasets are pertinent to this study: single family res idential properties in Al achua County and existing parks within the Gainesville city limits Property data was retrieved from the A ter appra isal file Neighborhoods are defined by census blocks using 2010 Census data. This d ata was j oined with property data to create one layer of properties and t heir respective neighborhoods. P arcel data was modified to include only parcels located in Gainesville, with the property use of single family re sidential (SFR), reflecting qu alified sales, with and with a lot size greater than 10 acres. According to reflective of a dollar amount relative to t he assessed mark et value. For example, a sale between family members which results in a sale price lower than the market value would be considered unqualified. Properties classified as unqualified sales were removed because initial analysis was focused on sales price fo r one particular year. H oweve r, upon conducting fur ther research, it was decided that using sales price would present a challenge in data analysis because of the variations in the real estate market over different years. As a result, properties from all sales years were used, and market value was used as the dependent variable. Market value is determined by the Alachua County Property Appraiser using a computer assisted mass appraisal (CAMA) syste m, and includes the consideration of the cost to reproduce the property, the ability of the property to earn income, and sales prices of other similar properties in the area. The average difference between the sale price and market value in this dataset was $761.66.
42 Data regarding parks was retrieved from th e Alachua County G rowth Management Geographic Information System ( GIS ) database. This database contains a dataset of 66 parks located within the city limits of Gainesville. They are owned and managed by public and private organizations, including Alachua County, the City of Gainesville State of Florida, Boys Club of Alachua County, Girls Club of Alachua County, Gainesville Housing Authority, and the Florida Audubon Society. The largest park in Gainesville is Gum Root Park, a community natural resource park owned by the City of Gainesv a neighborhood child recreation park owned by the City of Gainesville, and is .10 acres. Alachua County s eparates parks into three classifications based on the region that they serve : n e ighbor hood, community, and regional. The County wide Recreation Master Plan defines n eighborhood parks as being small parks usually between 5 and 9 acres and if recreation based, providing resources on a small scale. This can include practice areas, green spac es, and playing courts. They are typically located in close proximity to elementary sc hools and are designed to be accessible by a short bicycle or car commute. Community parks are designed to provide service on a wider scale. If the park provides recre ational resources, it is for activities such as team sport s They can include facilities such as athletic fields, swimming pools, and recreation centers. They are usually between 20 and 50 acres in size and are typically located near schools or other co mmunity services such as libraries. In addition to these classifications, parks are placed in the following categories based on their use and resources : child recreation, leisure, mixed recreation, natural resources, public use facilities, and sports rec reation. The category with the highest frequency and the greatest number of acres
43 is natural resources acreage Table 3 1. Park categories in study area. Park Type Number of Parks Acreage Child Recreation 13 198.10 Leisure 8 135.50 Mixed Recreation 1 4 347.40 Mixed Resources 1 102.80 Natural Resources 17 1,524.20 Public Use Facilities 7 125.10 Sports Recreation 6 102.70 Total 66 2,433 Regression Analysis Two methods were utilized t o perform regression analysis: ordinary least squares (OLS) and geographica lly weighted regression (GWR). Applying the principles established in reviewed literature, OLS was used as the first approach to answer the research question. The following model is used as a basis for analysis : Hp i = f(S i N i L i ) (3 1) Where: Hp i = Market Value of the i th prope rt y S i = Structural attributes N i = Neighborhood Attributes L i = Locational Attributes Market value was used as the dependent variable in this m odel. Independent variables include d structural attributes, neighborhood a ttributes, and locational attributes, specifically park attributes Structural attributes include d parcel size in acres, building
44 size in square feet, building age in years, number of bedrooms, number of bathrooms, and building stories. Parcel size, square footage, number of bedrooms, number of bedrooms, and number of stories were expected to have a positive correlation with market value. Age was expected to have a negative correl ation. Neighborhood variables include d population den sity and percentage minority Population density was calculated as the number of peo ple in a c ensus block divided by the number of acres This variable was expected to exhibit a positive correlatio n c oefficient because of Gainesville being the location of the University of Florida which employs residents throughout Alachua County. It is more desirable to live in the city and be in closer proximity to the UF campus ; therefore demand is higher resulti ng in higher prices. 279 people per square mile compared to of 2,028.5 per square mile. C ensus data was limited to demo graphic information at the time of this study ; therefore perc ent minority was used as a proxy for low income. Because of this, a negative correlation coefficient was expected for this variable. ( See figures 3 2 and 3 3 for maps illustratin g population density and percent minority variables ) D istance to the neare st park was used as the locational variable. Additional locational variables were considered, including type of the nearest park, size of the nearest park, and distance to the University of Florida campus. Results of this alternative model were compared to initial results in terms of the degree of spatial autocorrelation and the observed level of reliability of the models. After com paring results t he final model used for analysis of specific park locations included only distance to park a s a locational variable.
45 Table 3 2. Variables used in regression analysis. Variable Name Variable Description Variable Type Expected Relationship ACRES_CALC Parcel size in acres Independent + SQFT Building size in square feet Independent + AGE Building age in years Independent BEDS Number of bedrooms Independent + BATHS Number of bathrooms Independent + STORIES Building stories Independent + DENPOP2010 Neighborhood population density Independent + PCT_MNRTY Percentage minority in census block Independent P ARK_DIST Distance to nearest park Independent MKTVAL Assessed Market Value Dependent N/A Three distances were considered for analysis: 500ft, mile, and mile. Crompton (2005) sug gests that the mile distance has been generally accepted as being equivalent to a 5 minute walk Distance in Arc GIS, which determines the distance from the nearest feature that i s specified (in this case, distance was calculated between parcels and the nearest park within the spe cified distances ). Considering the maximum search distance of mile, there were 9,317 properties in the dataset (see table 3 3) Table 3 3. Descriptive statistics of dataset. n = 9 317 Minimum Maximum Mean Std. Deviation Acres .10 11.12 .32 .30 Market Value 79 00 805900 138880.12 64949.22 Sqft 399 10282 2141.41 850.04 Age 0 110 41.39 49.12 Beds 1 5 3.08 .60 Baths 1 7 1.91 .56 Population Density 0 62.96 5.60 3.14 Percent Minority 0 100 32.12 24.55 Distance to Nearest Park (ft) 26.16 2639.87 1355.25 707.03
46 As noted by Bitter (2007) and Wheeler and Pez (2010) t he use of ordinary least squares presents a problem when cons idering spatial relationships because of spatial heterogeneity. This method assigns a global correlation coefficient to each variable which is essentially an average of the coefficien ts of all points in the dataset In order to account for spatial variation, geographically weighted regres sion was utilized. This method expands upon the ordinary least squares model by providing coefficients for e ach variable at each location. Because the relationship between variables fluctuates over space, three park locations were chosen for further analysis. The first two parks were chosen based on the fact that they were categorized as nat ural and leisure resources, categories that would be expected to have a low er comparative level of recreation use and in turn a lower level of ne gative externalities such as noise or privacy reduction In addition, upon reviewing results of geographicall y weighted regression analysis, both parks exhibited a variation in positive and negative coefficients regarding park distance. The third location was chosen based on the fact that it is a public use fac ility with the primary goal of providing for child c are and recreation, as well as the fact that the majority of properties located within mile of a park exhibited positive coefficients in regards to the park distance variable. It was expected that the first two parks would exhibit a variation of positiv e and negative correlation coefficients, and that a majority of correlation coefficients for the third location would be positive (indicating that properties do not b enefit from living adjacent to that facility).
47 Figure 3 2. Map of study area minor ity percentages by census block.
48 Figure 3 3. Map of study area population density by census block.
49 CHAPTER 4 F INDINGS AND RESULTS Ordinary Least Squares (OLS) The primary concern of this study was the impact of proximity to parks; therefore, the fir st regression analysis that was performed included distance to the nearest park as the locational explanatory variable Using the first search distance of 500ft resulted in a dataset of 1,290 properties with an average market value of $139,318. Average d istance t o the nearest park was about 295 ft (see Table 4 1). Table 4 1. De scriptive statistics of properties located within 500ft of a park. n = 1290 R esults of OLS regression analysis indicate d an adjusted R squared value of .82, mean ing that 82% of market value could be explained by the variables used in this model. Number of bedrooms, bathrooms, square footage, po pulation density, and percent minority were found to be statistically significa nt at the .01 level. T he variable with the most influence in this model was square feet, which had a t statistic of 36.43. According to these results, every square foot added to a prop erty equates to an added value of $58.88 Number of s tories, age, and park distance were not found to be Minimum Maximum Mean Std. Deviation Acres .10 3.44 .35 .33 Market Value 7,900 709,500 139,318 75894.40 Sqft 44 0 9105 2,150 1000 Age 1 110 4 4 20.33 Beds 0 5 3 .67 Baths 0 5 1.87 62 Population Density 0 31.22 5.50 3.65 Percent Minority 0 100 34.32 27.11 Distance to Nearest Park (ft) 26.15 500 294.50 133.73
50 statistically significant in this case; the probabilities for these variables w ere all above .05 (see table 4 2 ). Table 4 2 OLS r esults for properties within 500ft of a park. Variable Coefficient Probability T Stat istic ACRES_CALC 1702.46 .60 .84 BEDS 4911.80 .00* 2.96 BATHS 17380.86 .00* 8.46 STORIES 3492.98 .12 1.70 SQFT_1 58.88 .00* 39.43 DENPOP2010 784.59 .00* 3.03 PCT_MNRTY 375.03 .00* 10.33 AGE 17.25 .15 1.45 PARK_DIST 0.70 .99 0.01 Adjuste d R 2 = .82, n=1290 *Indicates statistical significance. (See Appendix for full results.) The second OLS regression analysis performed included properties within a mile search radius of a park. This dataset included 4,563 parcels with market values ra nging from $7,900 to $805,900, and an a verage market value of $139,318 Average distance to the nearest park for these properties was 736.55. (See table 4 3). Table 4 3 Descriptive statistics of properties located within mile of a park. n = 4563 Minimum Maxim um Mean Std. Deviation Acres .10 5.27 .33 29 Market Value 7,900 805,900 1 39,318.33 71410.00 Sqft 440 10,282 2138.80 932.21 Age 0 110 43 18.16 Beds 0 5 3.07 63 Baths 0 7 1.88 .60 Population Density 0 31.22 5.63 3.13 Percent Minority 0 1 00 33.03 2 6.90 Distance to Nearest Park (ft) 26.16 1320.54 736.55 348.62
51 The adjusted R squared value for this mode l was the same as the previous model, .82. All variables except for park distance were found to be statist ically significant All but the age variable were significant at the .01 level. The variable with the greatest influence in this case was again square feet with a t statistic of 99.93; e very additional square foot in a house was worth an additional $59.15 consistent with resu lts at th e 500ft search radius. However, relationships between structural variables a nd market value were not all con sistent with expected outcomes. N umber of bedrooms exhibited a negative correlation coefficient, sugge sting that more bedrooms equated to a lower property value This was a sign that the model may not be completely reliable. Table 4 4 OLS r esults for properties within miles of a park. Variable Coefficient Probability T Statistic ACRES_CALC 6218.71 .00* 4.30 BEDS 2696.95 .00* 4.85 BATHS 11 280.32 .00* 13.03 STORIES 3707.83 .00* 6.61 SQFT_1 59.15 .00* 99.93 DENPOP2010 1237.18 .00* 9.98 PCT_MNRTY 389.14 .00* 27.65 AGE 20.44 .04* 3.16 PARK_DIST .98 .47 2.26 Adjusted R 2 = .82, n=4563 *Indicates statistical significance. (See Appendi x for full results.) The final OLS regression analysis performed included properties within a mile search radius of the nearest park. This dataset included 9,317 parcels with market values rang ing from $7,900 to $805,900, and a mean market value of $13 8,880.12. Average distance to the nearest park for these properties was 736.55 feet (see figure 4 5).
52 Table 4 5 Descriptive statistic s of properties located within mile of a park. n = 9317 Results indicated an adjusted R squared value of .80, suggesting that 80% of the relationship between variables was explained by the model. All explanatory variables exhibited statistical significance at this level. Park distance was signifi cant at the .05 level, and all other variables were significant at the .01 level. The variable with the greatest predictive power in this model was again square feet, with a t statistic of 99.3. The result for this variable was as expected; the positive coefficient of 59 indicated that for every square foot, property value goes up by $59. Table 4 6 OLS r esults for properties within mile of a park. Variable Coefficient Probability T Statistic ACRES_CALC 5020.93 .00* 4.30 BEDS 3125.35 .00* 4.85 BATHS 9879.47 .00* 13.03 STORIES 5505.82 .00* 6.61 SQFT_1 59.30 .00* 99.93 DENPOP2010 1051.31 .00* 9.98 PCT_MNRTY 386.30 .00* 27.65 AGE 18.93 .00* 3.16 PARK_DIST 1.01 .02* 2.26 Adjusted R 2 = .80, n=9317 *Indicates statistical significance. (S ee Appendix for full results.) Minimum Maximum Mean Std. Deviation Acres .10 11.12 .32 .30 Market Value 7,900 805,900 138,880.12 64945.72 Sqft 399 10,282 2141.41 850 Age 0 110 4 0 18. 20 Beds 0 5 3.0 8 .6 0 Baths 0 7 1. 90 56 Population Density 0 63 5.57 3.14 Percent Minority 0 100 33.03 26.90 Distance to Nearest Park (ft) 26.16 1320.54 736.55 348. 62
53 Park distance was negatively correlated, though by a small amount. The co rrelation coefficient of 1.01 suggested that for every foot increase in distance from a park, market value dropped by about $1. Although all variabl es were indicated as being statistically significant, some relationship directions were not c onsistent with expected results. T he coefficient for number of bedrooms indicated a n inverse relationship consistent with results from the first two models Var iables with positive coefficient values, including parcel size, number of bathrooms, number of stories, square feet, and population density, were consistent with expected results. Negatively correlated variables were accurate to predictions as well. Age would be expected to relate negatively to market value, with value going down as age increases. Finally, u sing percent minority as an indicator of income levels, the resulting negative coefficients make sense; residents with lower income will live in ho me s with a lower market value. Results of the three tested models exhibited signs of unreliability. Firstly, p roven variable relationships were not shown to be true In the first model using a dataset of properties within 500ft of a park, parcel size and a ge of the structure were not indicated as being statistically significant; it can rationally be expected that both of these variables would be significant in determining market value Age can be expected to exhibit a negative correlation ; newer housing un its are generally viewed as being more desirable resulting in an added premium Additionally, parcel size normally exhibits a positive relationship with value; the greater the size, the greater the value. The second model, analyzing attributes of proper ties within mile of a park, did result in all structural and neighborhood variables being significant; however, there was again inconsistency between proven relationships and actual results. Furthermore a ll models resulted in a
54 negative correlation coe f f icient for number of bedrooms. In addition to questionable indicated that data was not randomly distributed in any of the three models. Z scores for standard deviat ions in the model exhibited high values, indicatin g that the data was clustered, and suggested that there was less than a 1% chance that the pattern could be a result of random chance. In other words, the mode l cannot be fully relied upon. In an attempt to define a better performing model which did not exhibit the aforementioned issues especially the issues of spatial autocorrelation three additional explanatory variables were included for analysis: distance to the University of Florida, park size, and park type Keeping consistent with the previous models, the dataset for properties within a mile search distance of a park were used. There were seven types of parks according to the Alachua County Growth Management database. Dummy variables were use d to represent these types. Because of the large number of categories, and the fact that some of them were closely related in terms of type, collinearity problems arose and these variables were not able to be used. R esults indicated the significance of a ll variables at the .01 level except for park distance, which was not found to be significant using this model (See t able 4 7 ) However, use of these variables did not improve upon model perfo rmance. Number of bedrooms still exhibited a negative correlat ion, raising concerns over the reliability of the model, and results of the again indicated that data was not normally distributed.
55 Table 4 7 OLS results for pro perties within mile of a park (alternative mod el) Variable Coefficient Probability T Statistic ACRES_CALC 5020.93 .00* 3.18 BEDS 3125.35 .00* 3.45 BATHS 9879.47 .00* 7.79 STORIES 5505.82 .00* 4.49 SQFT_1 59.30 .00* 46.02 DENPOP2010 1051.31 .00* 9.04 PCT_MNRTY 386.30 59 26.91 AGE 18.93 00* 8.32 PARK_DIST 1.01 .00 2.26 ACRES 44 .00* 6.09 UF_DIST 0.83 .00* 0 68 Adjusted R 2 = 80 n=9 317 Geographically Weighted Regression (GWR) Because results of OLS models exhibited signs of inconsistency and spatial autocorrelation a g eographi cally weighted regression analysis was utilized to account for spatial variants. GWR was performed using the variables that were identified as being significant in the previously discussed OLS models. Park distance was only indicated as being sign ificant when using the mile search radi us, therefore this was the dataset utilized for GWR. 748 of the parcels in this dataset exhibited correlation coefficients of 1.797693e+308 ; these were removed from the da taset, leaving 8 ,569 properties (s ee table 4 6).
56 Table 4 8 Descriptive statistics for revised dataset after GWR. n = 8,569 Use of this method resulted in a higher adjusted R squared value of .92, indicating that 92% of the relationship could be expla ined by the included explanatory variables. Considering mean values for the coefficients of each variable t he average coefficient relationships were as expected for structural a nd neighborhood attributes. All coefficients were positive, with the exception of percent minority which was expected to be negative. As for the locational variable, park distance, the average of the correlation coefficients was positive. Although the value was small, 1.60, these results were contrary to the hypothesis. However, the benefit of using GWR is the ability to identify relationships at different locations instead of applying an average to an entire geographic area Out of 8,569 total proper ties, 3, 246 or about 38%, exhibited a negative correlation wi th park distance. In other words, 38% of properties exhibited increased property values as a result of proximity to a park (see figure 4 1) Coefficients for park distance ranged between 43.8 3 and 52.70. Looking only at properties with negative correlation coefficients average distance was 1326ft, or about mile Using t his mean distance and the mean negative correlation coefficient of 7, Minimum Maximum Mean Std. Deviation Acres .10 11.12 .33 .31 Market Value 7,900 805,900 143,020 65751.44 Sqft 399 10,282 2,181.84 867. 41 Age 0 110 40 18.72 Beds 0 5 3.08 .01 Baths 0 7 1.92 .57 Population Density 0 63 5.42 3.13 Percent Minority 0 100 29.24 22.61 Distance to Nearest Park (ft) 26.15 2639.880 1339.58 704.67
57 the average added value of parks to properties can be estimated at $9,289. Added value to properties with a positive correlation coefficient on average, was $6.80 per foot; an average of $9,159. However, as exhibited by results of this model, averages cannot be fully relied upon; therefore further ana lysis of specific geographic locations was needed. In an effort to defin e the best representation of the relationship between variables the alternative model including the variables of size of park and distance to the University of Florida was also anal yzed using geographically weighted reg ression. Dummy variables for type of park were not included as they presented problems of collinearity. This analysis resulted in an adjusted R squared value of .78, and a sigma value of 38 345.23. The original mode l, which had an R squared value of .92 and a sigma value of 26,958.51, was judged to be the better performing model, and thus was used for the further analysis of specific locations. However, postulations can still be made regarding the relationship of th ese variables on market value. The relationship between proximity to the UF campus and market value likely involves multiple variables, an d could lead to a separate study of how distance to the campus impacts market value or rental prices Variables rela ting to the rental market no doubt have an impact o n housing values in locations which are in close proximity to the campus. The population occupying properties in close proximity to UF is largely characterized by young college students that are renting a partments, single family homes or condos many of which a re shared with roommates. This portion of the population is likely to pay hig her rents to live closer to UF for the added benefits of being able to walk to campus nearby restaurants, and other est ablishments. Results for the distance to UF explanatory
58 variable may not have accurately represented the relationship b ecause analysis only included single family homes, which leaves out a large portion of the dwelling units that are occupied by students, such as apartments or condos. The rental market in Gainesville is also unlike rental markets in cities which do not house major universities. M any properties with multiple bedrooms are rented under individual leases, charging a higher premium for each roommate, resulting in an overall higher rental price than would normally be seen. In addition, t he portion of the population that is not made up of students has different preferences. Based on personal knowledge of the area, a majority of non transitory residents do not live in the areas directly surrounding the UF campus. Although pre ferences vary, for those residents that are not affiliated with the university, as well as for some that are, it is more desirable to live farther from campus rather than in walking di stance or a similar proximity. This preference may be based on multiple factors, which are difficult to control for. T raffic levels around the campus can be a deterrent as well as the existence of many pedestrians an d bikers in the area. I n addition, the atmosphere of some areas may dissuade residents from locating there who have families and are thus seeking a more family oriented location, and have different amenity needs such as schools, daycares, etc. Furthermore residents who are old er or simply those that desire to live in a quiet location would likely choose other parts of Gainesville to reside.
59 Figure 4 1 Park distance correlation coefficients for properties within mile search radius.
60 As previously mentioned, in order to gain a better understanding of the varying degrees of impact that park location may have, three separate par ks of different type and size were identified for a more detailed analysis. The first park was Cofrin Nature Park, o w ned by the City of Gainesville. Co frin Nature Park is 30 acr e s, and is designated by Alachua County as a natural r esources park which serves the community region. Its amenities include a half mile long hiking trail as well as a playground, and seepage wetlands above Beville Heights Creek which support growths of plants and wildflowers (City of Gainesville, 2011). According to 2010 Census data, t he average pop u lation density for this area is 4.6 people per acre; an ave rage of 15% of the population was made up of minorities. 525 propert ies were located within a mile search radius of the park, with a mean value of $ 176 739.61 405 properties ( 77%) exhibited negative correlations between market value and park distance (see figure 4 3 ). Considering p roperties with negative correlations, o n average, for every foot increase in proximity to the park, $1 3.50 was added to market value. Using the average park distance of 1449.82, this equates to an average of $19,572.57 added value. Although negative cor relations are seen at varying distance s within the mile radius positi ve correlations start to become less prevalent around 12 00ft distance (see figure 4 2 ). Figures 4 3 and 4 4 illustrate the degree of variance among correlation coefficients for park distance. Properties north of 8 th Aven ue are of particula r importance in this location and provide a solid example of why specific locati ons should be further analyzed 8 th Avenue, which is to the south of the park, as well as Newberry Road, also to the south, create a division between proper ties that are located within the
61 mile search distance and indicate that the two areas should not be considered in the same way. Properties located to the south of the park do not have the same level of accessibility as those located on the northern side in terms of walkability ; residents of this area must either cross Newberry and/or 8 th Avenue, or drive to the park. As for properties to the north of 8 th Avenue, h ouses on the western side of the park are not located on the direct perimeter, but are buff ered by prop erties with a smaller parcel size, less than .10 acres (which is why these properties were not included in the dataset) (see figure 4 4 ). Following the net effects curve theory, the smaller properties which are direc tly adjacent to the park ac t as a buffer for nuisance variables such as noise or access points. Properties on the northern and eastern perimeter, on the other hand, do not have this barrier, and exhibit positive correlation coefficients. Additional locational variables may impact the relationship between variables in this case, such as the location of a daycare near the park, which provides services for different ages of children and likely results in increased noise as well. Figure 4 2. Park distance correlation coefficients C ofrin Nature Park.
62 Figure 4 3. Park distance correlation coefficients for properties within mile of Cofrin Nature Park.
63 Figure 4 4 Graduated p ark distance correlation coefficients for properties within mile of Cofrin Nature Park.
64 T he second park that was identified for further analysis was Springtree Park, located on the n orth west side of Gainesville. Owned by the City of Gainesville, it is an 11 acre neighborhood park and falls in the type category of natural resources Amenities located for p ublic use on the site include a sma ll playground and picnic area In addition, there are nature trails which pass through areas surrounding Three Lakes Creek, a tributary of Possum Creek (City of Gainesville, 2011) The a verage population density of this area is 5.6 people per acre. 874 properties were located within the mile search distance of the park, with an a verage market v alue of $ 124 560.64 Results for correlation coefficients in this model were as expected. Using mean values, structural at tributes including parcel size, number of bedroo ms, number of bathrooms, number of stories, square feet, and population density exhibited positive correlation coefficients. Average values for correlation coefficients of age, percentage minority, and di sta nce to parks were negative. Of the 874 properties, about 50% exhibited negative correlations between market value and park distan ce meaning that the value contributed by the park got smaller as distance increased, consistent with the hypothesis (see figu re 4 5). Correlation coefficients ranged from 19.15 to 16.23. The first notable observation in this case is the division of impact between properties located on the north and south side of 39 th Avenue. Positive and negative correlations are seen on the north side ; however almost all properties on the south side exhibit park distance coefficients with positive values Figure 4 6 shows the varying degrees of park distance coefficients. Similar to the previous example, it is suggested that the area nor th of 39 th Avenue should be the focus of this analysis as it is the location of Springtree neighborhoo d. Based on persona l knowledge of the area, 39 th Avenue, along with NW
65 34 th Street and NW 53 rd Avenue acts as an edge to the Spri ngtree neighborhood ; th ere is a clear mental boundary between this neighborhood and locations south of 39 th Avenue. Also of particular interest in this case is the point at which park distance coefficients become negative. A majority of properties on the perimeter of the park do not benefit positively from the park. Considering only the properties to the north of 39 th Ave nue 41 properties within 583ft of the park exhibited positive correlation coefficien t s for the park dist ance value As the net effects theory suggests, this could be a result of n egative extern alities from the park In this case, living on the edge of the park provide s for easier access ; many residents can walk from their backyard directly into the park area. m ay result in issues such as unwanted noise and reduced privacy levels There are not specific access points to this park; patrons can access it by walking through the backyards of residents on the perimeter. In addition to the characteristics of this par k, some of the relationship may be explained by the existence of a small field used for outdoor activities by the neighboring property, a religious establis hment located to the east of the park Based on personal observation, when this activity field is i n use, the noise generated is highly Residents living in close proximity but no t directly on the perimeter still experience the positive externalities of easy accessibility to natural amenities, b ut are not subject to the negative ext ernalities that affect properties directly abutting the park Using the average correlation coefficient and park distances for negatively correlated properties, 4.7 and 1341.20 respectively, the average estimated add ed value of parks to a property in this area is $6,303.64.
66 Figure 4 5 Park distance correlation coefficients for properties within mile of Springtree Park.
67 Figure 4 6 Graduated p ark distance correlation coefficients for properties within mile o f Springtree Park.
68 The third park chosen for further analysis was the Boys Club of Alachua County a 6.4 acre public facility on the south east side of Gainesville. This facility was chosen as a comparison to the aforementioned natural resource and leisu re activity parks because of its positive correlation coefficients for properties in the mile search area. 1 64 prope rties were located within the search distance with a me an value of $78,02 3 The relationship between market valu e and park distance was as expected in this case; 82% of properties exhibited a positive correlation with park distance (see figures 4 7 and 4 8 ) I n other words they were not positively impacted by their location proximate to the park. Again using the net effects principle as a basis, this may be the result negative externalities While positive externalities are produced by the preservation of open space, p resents nuisance issue s including unwanted noise, traffic, or privacy issues fo r residents living in houses ar ound it. Comparing the results of all three park analyses provides support for the hypothesis that the degree of negative externalities does affe ct the impact of the park distance variable. While the first two parks do have a recreation component, their primary characteristic is the preservation of natural resources. In addition, they are both larger, Springtree Park being 11 acres and Cofrin Nat ure Park being 30 acres. The extra open space containing trees and vegetation may act as a buffer to the recreational activities that do exist on these properties. In contrast, the Boys Club is only 6 acres, and is characterized largely by cleared open s pace, providing less opportunity for negative externalities to be buffered. Finally, the nature of the park as a non profit owned facility,
69 as opposed to a publicly owned and accessible park like the first two examples likely makes a difference in the de sirability of living near it. Figure 4 7 Park distance correlation coefficients for properties within mile of the Boys Club of Alachua County.
70 Figure 4 8 Graduate park distance correlation coefficients for properties within mile of the Boys Club of Alachua County.
71 CHAPTER 5 DISCUSSIONS AND CONCLUSIONS Discussions It has been established that housing is a m ulti attribute good. The total value of a house is the result of the values of the bundle of attributes that make it up. Review of literat ure has shown that t he attributes with the largest contribution to price are usually structural characteristics, such as square footage, acreage, and number of bedrooms. Neighborhood characteristics such as income levels and population density have been s hown to be significant contributing factors as well. Locational or environmental factors, such as proximity to amenities like parks, while not as significant, have also been shown to contribute to total value. Interest in determining a quantifiable value of parks gained and Hendon in 1967. Their study utilized simple regression analysis to model the relationship between park proximity and the value of housing and land; housing and neighborhood characteristics were not taken into considerat ion. Their results showed that land values alone were positively impacted by the pro ximity of a neighborhood park. The lack of a method to account for housing heterogeneity was cited as reasoning for their results. As these methods evolved, results which included a more complex analysis of housing values began to surface; many of which suggested that not only land, but also housing values, are positively impacted by proximity to park s. Morancho (2003) and Bolitzer and Netusil (2000 ), among several others, provide evidence supporting this hypothesis Not only have more modern studies provided ev idence to support the theory that proximity to parks increases housing values, they have a lso suggested that the value which is seen is dependent upo n the size an d /or
72 type of park. The reviewed studies, along with others cited, have used regression analysis methods such as ordinary least squares (OLS) and variations thereof, to provide these result s. As method s continue to evolve, so too do the opportunities to keep improving upon these models. Ge ographically w eighted regression (GWR) has been presented as a method which addresses a problem that stationary coefficient models cannot spatial heterogenei ty. Rather than apply o ne coefficient to each variable at a global level, geographically weighted regression applies a correlation coeffi cient to every feature in a dataset This method can address model issues such as omitted variables, and otherwise un identifiable spatial issues suggested as a result The primary concern of t his study was to test the hypothesis that proximity to a park h as a positive impact on housing value. Using foundations laid by reviewed literature, this relationship was tested using the market value of single family homes in Gainesville, FL As in the cited studies, variables included in analysis reflected structural, neighborhood, and locational attributes. Results of ordinar y least squares regressions considering properties within 500ft and mile of a park indicated that there was not a significant statistical correlation between market value and park distance. Using a search distance of mile s howed that not only was ther e a significant relationship between the two variables, but also that there was although small, a n inverse relationship between housing value and park d istance. This suggests that as distance from a park increases market value decreases. The adjusted R squared value of .80 was strong; however, i ndications of unreliability were present in results; some variables were inconsistent with expected findings. For instance, in all ordinary
73 least squares models, the correlation coefficient for number of bedroom s was negative. This is contrary to the logical assumption that number of bedrooms adds to housing value. In addition, a test of s patial a for all three models indicated that the data was not randomly disbursed. Thus, further r esearch was needed to achiev e more reliable model results. Utilization of geographically weighted regression improved upon the OLS model by assign ing a coefficient at every feature in the dataset allowing for further analysis of the relationship between variables at specific locations. The results showed that relationships vary over space; where OL S assigns one coefficient for a variable at all locations, GWR shows that the coefficients for these variables can actually be negative in some places, and pos itive in others. Considering mean values for the coefficients of each variable showed that the average coefficient relationships were as expected for structur al and neighborhood attributes. Al l coefficients were positive, with the exception of percentage minority. Using this variable as a proxy for income, the resulting negative coefficients make sense; residents with lower income would be expected to live in ho mes with a lower market value. Three parks were chosen for further analysis in order to take into account different types and characteristics. Resulting coefficients showed that while park distance can have a positive impact on some properties, it is not uniform over the entire study area. In regards to Cofrin Nature Park and Springtree Park, pr operties on the perimeter are negatively impacted by the parks location, while properties a short distance away are positively impacted The character and size of parks likely does affect the economic impact to proximate properties because of the externali ties they present. Recrea tion based
74 parks, especially community parks that provide for recreation activities on a larger scale than a single basketball court or small field present a higher degree of negative externalities due to noise, access point traf fic, and reduced privacy levels. Properties located directly adjacent to parks that provide for team sports, child recreation, or public use facilities may be especially s ubject to these disadvantages. Results of the analysis of parks within mile of th Homes located directly adjacent to l arger natural resource or leisure activity based parks are more likely to be positively impacted as a result of the positive externalities such as stormwater control, air quality, a nd even the reduction of noise pollution from parks, which can act as a buffer between properties and roads, commercial activity, or otherwise noise producing elements. Smaller resource based parks may create a greater degree of negative externalities f or the properties located on the park perimeter because of thei r neighborhood level character. Using Springtree Park as an example, properties which are on the direct perimeter of a park with no designated access points are subject to the negative externa Limitations T here are several limitatio n s associate d with using regression models including the choice of which functional form is used, problems associated with collinearity and the di fficulty of includi ng all variables that may be pertinent In addition, the issue of spatial heterogeneity is important to consider, and should lead to the use of regression models that take spatial variants into account Reviewed literature did not acco unt for spatial variants, but instead utilized models that assumed homogeneity among properties. As seen in the r esults of this study, results va ry based on which method is used.
75 A vailabi lity of data was also a limitation. Census data fr om 2010 regardin g demographics were available ; however data relating to incom e and poverty levels were not. Percentage minority has been suggested as an acceptable proxy for income data; however, the utilization of actual income data may have resulted in a more accurate model. In addition, applying a specific model to multiple loca lities in general presents a limitation in itself because of the varying character a nd preferences of communities. ince a substantial amount of the population is made up of students attending the University of Florida or Santa Fe College, distance to those ca mpuses may be important to resid ents in some areas, but not others Students of the University of Florida may be more willing to pay a h igher rent to live close to the campus than employees who might prefer to live farther from the student housing areas surrounding it. In short, variable relationships cannot be globally summarized. Conclusions T he potential for parks and other permane ntly preserved open spaces to provide necessary ecosystem services in terms of environmental sustainability as well a s human health needs is notable Considerable evidence has a ffirm ed this fact. However, as with many environmental resources, the benefit is di fficult to define i f not translated into economic terms; therein lies the problem. How can a non market good ( such as access to r ecreational opportunities or increased air quality ) be quantified? By utilizing hedonic pricing models, the value of pa rks can be estimated as the added value to residential properties Several studies have shown that there is a significant statistical correlation between the distance from a home to a park or open space and housing value sometimes dependent upon various factors such as size or the type of
76 park. This study addr esses the issue by asking the following research question s : do parks have an economic impact on proximate residential property values? If so, is the impact positive or negative? Results indicate t hat parks do have an economic impact on proximate properties The answer to the question of whether the impact is positive or negative is not as straightforward. 38% of properties in the dataset of single family residential homes utilized for this study exhibited a n inverse relationship between park distance and market v alue indicating that as distance increases, market value decrease s Further, 77% of properties within mile of Cofrin Nature Park and 50% of properties within mile of Springtree Park exhibit this negative correlation coefficient for park distance However, 82% of properties within mile of the Boys Club were positively correlated with distance, indicating that these properties did not benefit from their location near this resource. Additional consideration of the character of these parks and the residential areas that surround them provides a better unde rstanding of the relationship. Which method should be used to estimate the value of a specific attribute in terms of housing price is debatable. Numerous studies utilizing linear techniques have concluded that homes located near parks tend to be positive ly influenced by proximity; however, the challenge with these models lies in the f act that a majority of the models used do not acc ount for spatial heterogeneity. Review of literature for this study found no exampl es of the use of methods which take into account spatial variations The outcomes presented here show the degree to which results can vary depending on which method is use d. When estimating the relationship between park distance and residential properties in Gainesv ille using the ordinary least squares technique there
77 was an inverse relationship at one distance, mile, but no significant relationship was found using data sets of properties located within a smalle r distance. Utilization of the geographically weighted regression method for the sa me dataset provided compelling evidence that the relationship between independent and dependent variables, specifically a location al variable in this case, can be positive o r negative These r esults suggest t hat the direction of correlation between park distance and market v alue can vary based on the type an d characteristics of the park including variables such as how much noise is generated from the various amenities that the park provides as well as the way in which the park is accessed. Further examination of three specific parks in Gainesville provided evidence that natural resource parks are more desirable to live near than parks with public use func tions that lead to increased negative externalities. The most important conclusion drawn from this study is that locational attributes cannot be assessed o n a global scale, but rather should be investigated on a case by case bas is, taking into account variations in spatial relationships.
78 APPENDIX A RESULTS OF ORDINARY LEAST SQUARES REGRESSION ANALYSIS Figure A 1. Ordinary Least Squares results, 500ft search radius Figure A 2. Ordinary Least Squares results, m ile search radius.
79 Figure A 3 Ordinary Least Squares results, mile search radius Figure A 4. Ordinary least squares results, using alternative explanatory variables, mile search radius
80 APPENDIX B RESULTS OF GEOGRAPHI CALLY WEIGHTED REGRE SSION ANAL YSIS Figure B 1. Geographically Weighted Regression Results for properties within a mile search radius
81 A B C
82 D E F
83 G H Figure B 2. Geographically Weighted Regression Results Frequency Distributions
84 Figure B 3. G eograph ically Weighted Regression Results mile search radius for alternative model including additional explanatory variables.
85 APPENDIX C PARKS IN GAINESVILLE FL NAME ACRES TYPE REGION OWNER ADDRESS Girl Scouts Kiwanis Park 2.30 CR Nhd City of Gainesville 810 NW 8th St Mini Park #02 0. 50 CR Nhd City of Gainesville 832 SE 9th St Mini Park #04 0.20 CR Nhd City of Gainesville 424 NW 6th Ave Mini Park #05 / Barbara Higgins Park 0.60 CR Nhd City of Gainesville 1360 SE 2nd St Mini Park #06 0.60 CR Nhd City of Gainesville 2003 NW 32nd Pl M ini Park #07 0.20 CR Nhd City of Gainesville 318 SW 7th Pl Mini Park #09 0.10 CR Nhd City of Gainesville 820 NW 4th Ave Mini Park #3 / Pleasant Park 1.10 CR Nhd City of Gainesville 510 NW 2nd St Mini Park #8 0.20 CR Nhd City of Gainesville 1645 NE 8th A ve Oak Hill Park 0.30 CR Nhd City of Gainesville 4100 NW 9th St Roper Park 1.60 CR Nhd City of Gainesville 401 NE 2nd St San Felasco Park 185.50 CR Nhd Alachua County 6400 NW 43rd Way Smokey Bear Park 4.90 CR Nhd State of Florida 2327 NE 15th St Cedar Grove Park 1.90 LA Nhd City of Gainesville 1200 Block NE 22nd St Chapman's Pond at Kanapaha Park 42.50 LA Cmnty City of Gainesville 7100 SW 41st Pl Lynch Park 1.40 LA Nhd City of Gainesville 450 S Main St Possum Creek Park 74.70 LA Cmnty City of Gaines ville 4009 NW 53rd Av Springhill Nhd Park 3.60 LA Nhd City of Gainesville 900 Block SE 4th Ave Sweetwater Park 3.20 LA Cmnty City of Gainesville 501 E University Ave Thomas Center Gardens 5.70 LA Cmnty City of Gainesville 306 NE 6th Ave University Park Arboretum 2.50 LA Nhd State of Florida 124 NW 23rd St A.N.N.E (Mini Park) 0.90 MR Nhd City of Gainesville 6310 NW 28th Ter Fred Cone Park @ Eastside Recreation Center 91.40 MR Cmnty City of Gainesville 2841 E University Ave Green Acres Park 38.80 MR Nh d City of Gainesville 3704 SW 8th Ave
86 NAME ACRES TYPE REGION OWNER ADDRESS Mini Park #01 1.00 MR Nhd City of Gainesville 1504 NE 4thAve NE 31st Ave Park 2.20 MR Cmnty City of Gainesville 1710 NE 31st Ave NE Cmnty Center 0.50 MR Cmnty City of Gainesvill e 1701 NW 8th Av Northside Park 31.50 MR Cmnty City of Gainesville 5701 NW 34th St Porter's Cmnty Center and Park 1.30 MR Cmnty City of Gainesville 512 SW 6th Ave Rosa Williams Park 0.90 MR Nhd City of Gainesville 524 NW 1st St T.B. McPherson Park 14.6 0 MR Cmnty City of Gainesville 1717 SE 15th St Tumblin Creek Park 9.60 MR Cmnty City of Gainesville 600 SW Depot Rd Veteran's Park at Kanapaha 23.00 MR Cmnty Alachua County 7400 SW 41st Pl Westside Park 26.10 MR Cmnty City of Gainesville 1001 NW 34th St Woodlawn Park 2.80 MR Cmnty Gainesville Housing Authority 1900 SE 4th St Boulware Springs NP 102.80 MRE Cmnty City of Gainesville 3500 SE 15th St Alfred Ring Park 20.60 NR Cmnty City of Gainesville 2002 NW 16th Ave Bivens Arm Nature Park 80.30 NR Cmnt y City of Gainesville 3650 S Main St Clear Lake Nature Park 12.90 NR Nhd City of Gainesville 5480 SW 1st Ave Cofrin Nature Center 30.30 NR Cmnty City of Gainesville 4810 NW 8th Ave Colclough Pond Audubon Sanctuary 35.40 NR Cmnty Florida Audubon Society 2315 S Main St Colclough Pond Nature Park 4.80 NR Cmnty City of Gainesville 2315 S Main St Devil's Millhopper State Park 69.40 NR Rgnl State of Florida 4732 NW 53rd Ave Gum Root Park 370.10 NR Cmnty City of Gainesville 7300 NE 27th Ave Hawthorne Rail T rail 195.60 NR Rgnl State of Florida 3500 SE 15th St John Mahon Nature Park 8.10 NR Nhd City of Gainesville 4300 W Newberry Rd Loblolly Woods Nature Park 128.00 NR Cmnty City of Gainesville 3315 NW 5th Av Morningside Nature Center 275.30 NR Cmnty City o f Gainesville 3540 E University Ave NW 29th Road Nature Park 5.60 NR Nhd City of Gainesville 1502 NW 29th Rd Possum Creek Preservation 11.50 NR Cmnty City of Gainesville 2219 NW 34th St Split Rock Conservation Area 240.10 NR Cmnty City of Gainesville SW 20th Ave
87 NAME ACRES TYPE REGION OWNER ADDRESS Springtree Park 10.70 NR Nhd City of Gainesville 2700 NW 39th Ave Terwilliger Pond 25.50 NR Nhd City of Gainesville 460 SW 62nd Blvd Alachua County Fairgrounds 103.90 PU Rgnl Alachua County 2900 NE 39th Av e Boys Club NW 6.80 PU Cmnty Boys Club of Alachua County 2601 NW 51st St Boys Club SE 6.40 PU Cmnty Boys Club of Alachua County 1100 SE 17th Dr Cmnty Plaza 1.20 PU Cmnty City of Gainesville 111 E University Ave Girls Club of Alachua County 5.10 PU Cmnt y Girls Club of Alachua County 2101 NW 39th Ave Sharmie Ffar Park 0.70 PU Cmnty City of Gainesville 925 NW 4th Pl Thelma Boltin center 1.00 PU Cmnty City of Gainesville 516 NE 2nd Ave Citizen's Field 7.00 SR Cmnty City of Gainesville 1000 Waldo Rd Fore st Park 26.60 SR Cmnty Alachua County 4501 SW 20th Ave Greentree Park / Kiwanis Challenge Playground 21.60 SR Cmnty City of Gainesville 1901 NW 39th Ave MLK Recreation Center 25.10 SR Cmnty City of Gainesville 1028 NE 14th St Mini Park #10 / Forest Pine s Park 0.20 SR Nhd Gainesville Housing Authority 1110 NE 25th St Northeast Park 22.20 SR Cmnty City of Gainesville 501 NE 16th Ave CR = Child Recreation LA = Leisure Activities MR = Mixed Recreation MRE = Mixed Resources NR = Natural Resources PU = Public Use Facilities SR = Sports Recreation
88 LIST OF REFERENCES Alachua County. (2010). Countywide recreation master plan. Retrieved 10/7, 2011, from http://www.alachuacounty.us/Depts/PW/Documents/parks/phaseI. pdf Anderson, J. E. (1985). On testing the convexity of hedonic price functions. Journal of Urban Economics, 18 (3), 334 337. Anderson, S. T., & West, S. E. (2006). Open space, residenti al property values, and spatial context. Regional Science and Urban Eco nomics, 36 (6), 773 789. Baum, F. (1999). Social capital and health: Implications for health in rural australia. presented at the Proceedings of the Fifth National Rural Health Conference, 96 109. orpo rating spatial variation in housing attribute prices: A comparison of geographica lly weighted regression and the spatial expansion method. Journal of Geographical Systems, 9 (1), 7 27. Bolitzer, B., & Netusil, N. R. (2000). The impact of op en spaces on prop erty values in Portland, O regon. Journal of Environmental Management, 59 (3), 185 193. Brigham, E. F. (1965). The determinants of residential land values. Land Economics, 41 (4), pp. 325 334. Retrieved from http://www.jstor.org/stable/3144665 Chiesura, A. (2 004). The role of urban parks for the sustainable city. Landscape and Urban Planning, 68 (1), 129 138. City of Gainesville. Parks Facilities 2011 from http://www.cityofgainesville.org/tabid/182/Default.aspx Commoner, B. (1971). The closing circle: Nature, man, and technology. New York, 141 265 268. Costanza, R., d'Arge, R., De Groot, R., Farber, S., Grasso, M., Hannon, B., et al. (1997). The value of the world's ecosystem services and natural capital. Nature 387(6630), 253 260. Craven, B., & Islam, S. M. N. (2011). ORDINAR Y LEAST SQUARES REGRESSION. The SAGE Dictionary of Quantitative Management Research, 224 228. Crompton, J. (2000). The impact of parks and open space on property values and the property tax base Natl Recreation & Park Assn. Crompton, J. (2005). The impact of parks on property values: Empirical evidence from the past two decades in the united states. Managing Leisure, 10(4), 203 218.
89 Driver, B.L., D. Dustin, T. Baltic, G. Elsner, and G.L. Peterson. 1996. Nature and the Human Spirit: Towar d an Expanded Land Management Ethic. State College, Pa.: Venture Freeman, A. M. (2003). The measurement of env ironmental and resource values: Theory and methods RFF press. Geoghegan, J. (2002). The value of open spaces in residential land use. Land use P olicy, 19 (1), 91 98. Goodman, E., & Whitaker, R. C. (2002). A prospective study o f the role of depression in the development and persistence of adolescent obesity. Pediatrics 110(3), 497. Irwin, E. G. (2002). The effects of open space on residential prope rty values. Land Economics, 78 (4), 465. Kitchen, J. W., & Hendon, W. S. (1967). L and values adjacent to an urbanm neighborhood park. Land Economics, 43 (3), pp. 357 360. Retrieved from http://www.jstor.org/stable/3145164 Knaap, G. (1998). The determinants of residential pr operty values: Implications for metropolitan planning. Journal of Planning Literature 12(3), 267 282. doi:10.1177/088541229801200301 Kuo, F. E., & Sullivan, W. C. (2001). Env ironment and crime in the inner city. Environment and Behavior, 33 (3), 343. Maller, C., Townsend, M., Brown, P., & St Leger, L. (2009 ). Healthy parks healthy people: The health benefits of contact with nature in a park context. Burwood (Australia): Faculty of Health and Behavio ural Sciences, Deakin University Miller, A. R. (2001) Valuing open space: Land economics and neighborhood parks Cambridge, MA, MIT Center for Real Estate. Milon, J. W., Gressel, J., & Mulkey, D. (1984). Hedonic amenity va luation and functional form specification. Land Economics, 60 (4), 378 387. Mo rancho, A. B. (2003). A hedonic valuation of urban green areas. Landscape and Urban Planning, 66 (1), 35 41. Nicholls, S., & Crompton, J. L. (2005). The impact of greenways on pr operty values: Evidence from austin, texas. Journal of Leisure Research 37(3), 321. Nicholls, S. (2002) Does open space pay? Measuring the impa cts of green spaces on property values and the property tax base, College Station, TX, Texas A&M University, PhD dissertation.
90 Office of Institutional Planning and Research. (2011). Universit y of florida fact book. Retrieved 10/6, 2011, from http://www.ir.ufl.edu/oirapps/factbooktest/enrollment/fb_enrollment.aspx Parsons, R (1991). The potenti al influences of en vironmental perception on human health. Journal of Environmental Psychology, 11 (1), 1 23. doi:10.1016/S0272 4944(05)80002 7 Polinsky, A. M., & Rubinfeld, D. L. (1975). Prop erty values and the benefits of environmental improvements: Theor y and measurement Institute of Public Policy Studies, University of Michigan. Racca, D. P., & Dhanju, A. (2006). Project report for prop erty Value/Desirability effects of bike paths adjacent to residential areas. Rosen, S. (1974). Hedonic prices and implic it markets: Product differentiation in pure competition. The Journal of Political Economy, 82 (1), 34 55. Santa Fe College. (2011). Santa fe college facts. Retrieved 10/6, 2011, from http://www.sfcollege.edu/facts/content/media/mugs/Facts%20 %20Fall%20Closi ng%202010.pdf Sundberg, R. (2002). Collinearity. Encyclopedia of Environmetrics Tyrvinen, L., & Vnnen, H. (1998). The economic v alue of urban forest amenities: An application of the contingent valuation method. Landscape and Urban Planning, 43 (1 3), 1 05 118. Wheeler, D. C., & Pez, A. (2010). Geographically weighted regression. Handbook of Applied Spatial Analysis, 461 486.
91 BIOGRAPHICAL SKETCH Brittany McMullen is currently pursuing a Master of Arts in Urban and Regional Planning as well as the i n terdisciplinary concentration and certificate in historic p reservation at the University of Florida. She has served as an intern for Al achua County Growth Management where she assisted in the evaluation and appraisal r eport process for upd ating the c ount o mprehensive p lan. She was also an inte rn for the city of Gainesville planning d epartmen t, specifically assisting with historic p reserva tion p lanning. She obtained a Bachelor of Arts in p olitical s cience, also f rom the University of Florida, with a minor in b usiness administration.