1# RECORD 79a14cd6c47cfe07a6fd907132f2ed86427b48312# TITLE: Are Homeowners Willing to Pay More for Access to Parks? Evidence from a Spatial Hedonic Study of the Cincinnati, Ohio, USA Park System3# YEAR: 2018.0 VENUE: Journal of Regional Analysis & Policy DOI: 10.68479/001c.38514# AUTHORS: Rainer Vom Hofe; Oana Mihaescu; Mary Lynne Boorn5# PAGES: 17 CANDIDATE TABLE PAGES (1-based): [7, 11, 13]6# ABSTRACT: This paper examines the impacts of the City of Cincinnati urban park system on residential property values using two local spatial hedonic model specifications: the spatial Durbin error model (SDEM) model, and the spatial lag of X model (SLX). Specifically, we examine how the distance to parks, the size of the nearest park, and the presence of park facilities for recreational activities, affect property values. Our results show that decreasing the distance to a park by one meter appreciates of the value of the average house in our sample by $3.44 to $6.27. A one hectare increase in the park size appreciates house values by $22.24 to $22.48 and the presence of park facilities for recreational activities depreciates house values by an estimated $4,612 to 5,300 for properties within a 500 meter radius to the nearest park.789===== PAGE 1 (score 12, numeric-paren 0, table-words 0, var-words 12) =====1011Journal of Regional Analysis & Policy1248(3): 66–8213Are Homeowners Willing to Pay More for Access to Parks?14Evidence from a Spatial Hedonic Study of the Cincinnati, Ohio,15USA Park System16Rainer vom Hofe17University of Cincinnati18Oana Mihaescu19HFI, Handelns Forskningsinstitut20Mary Lynne Boorn21University of Cincinnati22Received: 08/04/201723Accepted: 11/28/201724Abstract25This paper examines the impacts of the City of Cincinnati urban park system on residential property26values using two local spatial hedonic model specifications: the spatial Durbin error model (SDEM)27model, and the spatial lag of X model (SLX). Specifically, we examine how the distance to parks, the size28of the nearest park, and the presence of park facilities for recreational activities, affect property values.29Our results show that decreasing the distance to a park by one meter appreciates of the value of the30average house in our sample by $3.44 to $6.27. A one hectare increase in the park size appreciates house31values by $22.24 to $22.48 and the presence of park facilities for recreational activities depreciates house32values by an estimated $4,612 to 5,300 for properties within a 500 meter radius to the nearest park.33134Introduction35Cincinnati has enjoyed a long-standing tradition of public parks. The formation of the Cincinnati Board36of Park Commissioners in 1906 and the master plan for Cincinnati’s parks A Park System for the City of37Cincinnati by George Kessler in 1907 laid the foundation for what has become one of the nation’s most38recognized urban park systems. Cincinnati Parks is ranked 7 on the ParkScore® in 2015, published by39The Trust for Public Land.40One hundred years later, Cincinnati City Council adopted the Cincinnati41Parks 2007 Centennial Master Plan, which guides the Cincinnati Board of Park Commissioners’ mission “to42conserve, manage, sustain, and enhance parks’ natural and cultural resources and public greenspace for the43enjoyment, enlightenment and enrichment of the Cincinnati community” for the decades to come. Today,44the City of Cincinnati park system entails over 5,000 acres of city parklands, including five regional parks,4570 neighborhood parks, and 34 nature preserves, numerous natural areas and scenic overlooks, and a total of4650 miles of hiking and bridle trails. The park system amounts to approximately ten percent of Cincinnati’s47total land area, including the Cincinnati Zoo.48While voters supported the Cincinnati Board of Park Commissioners’ mission, they declined the slight49property tax increase necessary to maintain and improve the parks system. In the 2015 election, Cincinnati50voters overwhelmingly rejected a bill to add a one mill tax levy that would have raised an estimated $5 million51annually to improve the existing park system and expand bike trails in the area. Many local governments52face increasing financial pressure to balance budgets amidst rising expenses without increasing revenues.5366545556===== PAGE 7 (score 27, numeric-paren 0, table-words 2, var-words 16) =====5758Journal of Regional Analysis & Policy5948(3): 66–8260relevant hedonic literature, we included the following explanatory structural characteristics: the lot size61(sizeland), the size of the house in square meters (sizehouse), the number of bathrooms (bathrooms), the62age of the house (age), and the car capacity of the basement garage (garage) (Hess and Almeida, 2007;63Cebula, 2009; Landry and Hindsley, 2011). Following Bin et al. (2008), we added a variable denoting the64condition of the house, which includes seven categories, as defined by the Hamilton County Auditor’s Office:65very poor, poor, fair, average, good, very good, and excellent. We argue that the property condition is an66important contributor in determining its market value. We define the middle condition category (average) as67the baseline in our regression model. Consequently, we expect that houses in worse than average condition68(very poor, poor, or fair) will have lower values, while houses in better than average condition (good, very69good, or excellent) will have higher values. Definitions and summary statistics for the variables used in our70analysis are presented in Table 2.71Table 2: Variables: Definitions and Summary Statistics (n=36,167)72Definition73(Abbreviations in parenthesis)74Mean75Median76Minimum77Maximum78Market values of single-family residential79properties (P)80122,8528196,3008210,00083668,75084Structural characteristics of the property:85Size of the land, in hectares (sizeland)860.074870.056880.005892.01790Size of the building, in square meters (sizehouse)91162.94392146.6019322.29794547.75695Total number of bathrooms (bathrooms)961.709971.500980.000998.000100Age of the building, in years (age)10183.87810285.0001031.000104210.000105Basement garage capacity, # of cars (garage)1060.3351070.0001080.0001098.000110Condition of the house:111very poor (condvpoor)1120.0051130.0001140.0001151.000116poor (condpoor)1170.0161180.0001190.0001201.000121fair (condfair)1220.0961230.0001240.0001251.000126average (condaverage)1270.4551280.0001290.0001301.000131good (condgood)1320.3621330.0001340.0001351.000136very good (condvgood)1370.0651380.0001390.0001401.000141excellent (condexcellent)1420.0011430.0001440.0001451.000146Neighborhood characteristics:147Poverty rate (poverty)14826.46914924.0001502.50015182.500152% of people with bachelor degree (bachelor)15330.85915426.9001551.30015682.500157Crime rate (crime)15818.88615917.9161600.00016162.857162Public school performance indicator (school)16384.52116483.20016568.800166111.800167Distance to the closest park, in meter (parkdist)168248.930169245.4381700.000171499.997172Size of the closest park, in hectares (parksize)17345.08117411.5691750.018176576.135177Presence of park facility (parkfacility)1780.3751790.0001800.0001811.000182The poverty rate (poverty), crime rate (crime), and educational attainment (bachelor) are three com-183monly used neighborhood characteristics (Lynch and Rasmussen, 2001; Geoghegan, 2002). We use the 129184Census tracts in the City of Cincinnati as neighborhood proxy in our study.185Poverty rates and educa-186tional attainment were obtained from the 2010 US Census. The poverty rate is defined as the percentage187of the population below the poverty line; educational attainment refers to the percent of people with a188bachelor’s degree.189We define the crime rate as number of crimes / (population/1000).190Crime data are191publicly available for 2012/2013 from the City of Cincinnati police department.192Only property-related193crimes (i.e., aggravated burglary; inflicted harm; aggravated arson; breaking and entering; burglary; crim-194inal trespassing; fire weapon into habitat; property damage; safecracking; tampering with coin machine;195unauthorized use of property; and non-school vandalism) are considered for this analysis, since non-property196related crimes can take place randomly through the whole city and are thus not specifically related to197real-estate properties.198In addition, we include school district information to control for neighborhood199characteristics (Brasington and Haurin, 2006; Clapp et al., 2008; Dougherty et al., 2009).200Specifically,20172202203204===== PAGE 8 (score 14, numeric-paren 0, table-words 1, var-words 12) =====205206Journal of Regional Analysis & Policy20748(3): 66–82208we use a performance indicator for public schools, made available by the Ohio Department of Education209(http://reportcard.education.ohio.gov/Pages/default.aspx).210The variables of interest in our study are the Euclidean distances from each property to the closest park211(parkdist) in meter, the size of the closest park (parksize) in hectares, and the presence [absence] of park212facilities in the closest park (parkfacility). With the park facility binary variable, we distinguish between213natural green space and parks with built facilities, such as tennis, basketball, or soccer courts. Regarding214the distance measure (parkdist), we chose Euclidean over network distances for numerous reasons. First,215Euclidean distances are conceptually straight forward, are easy to compute, and have been successfully216used to capture location-related neighborhood characteristics in previous studies (Troy and Grove, 2008;217Conway et al., 2010). Earlier studies by Newell (1980) and O’Sullivan and Morrall (1996) reported that218network distances are approximately 1.20 to 1.23 times the Euclidean distance in an urban environment,219while Sander et al. (2010a) found the mean of the differences in distance to be as high as 48.4%. Further,220while Sander et al. (2010b) argue that road distances better reflect of actual travel distances to the nearest221park, we follow Eiser et al. (2007) who argue that Euclidean distances should be preferred to network (street)222distances when the perception of being in close proximity to the investigated feature is more important for223housing prices than accessibility. With respect to parks, either argument has some validity as some residents224perceive parks as a pleasant amenity. Without setting a foot into a park, they understand that parks build225communities and create a sense of place, thus contributing to a higher quality of life. Other residents see226parks as a welcome opportunity to walk their dogs, get some exercise through running, playing tennis or227basketball, or just socialize. While we do not know the specific percentages of active park users, we assume228that residents perceive nearby parks as a positive amenity. Last, the majority of parks can be accessed229literally from anywhere on foot. Consequently, it would be almost impossible to identify all park entrance230points.231According to the statistics reported in Table 2, the average house in our study has a market value of232$122,852. Our average house has 162.9 square meter, 1.71 baths, and is built on a 742 square meter lot.233The majority of houses in our study area are in either average (45.52%) or good condition (36.17%). Just234under 12% are in below average condition, while 6.65% are in above good condition. With respect to the235neighborhood characteristics, the average house is located in a Census tract with 26.5% of households living236below the poverty line, with 18.9 property-related crimes per 1,000 inhabitants, where 30.9% of the population237has at least a bachelor’s degree, and with an average school performance indicator of 84.5. Important for238our study is that the average house is located 248.9 meter away from the nearest park. This specific nearest239park is 45.1 hectares in size and there is a 37.5% chance that this nearest park includes some type of facility.2403.2241Empirical models242The centerpiece of this research is a hedonic price model, where potential buyers value the elements of the243bundle that compose a housing unit. The bundle is made up of structural and neighborhood characteristics,244including park characteristics. Following LeSage and Ha (2012) and Mihaescu and vom Hofe (2013), we245applied a Bayesian spatial Durbin error model (SDEM), defined as:246lnP = α + S ∗β + N ∗γ + W ∗S ∗θ + C ∗ϕ + u247(6)248u = λ ∗W ∗u + ϵ249ϵ ∼N(0, σ2250ϵ In)251where lnP is a vector of log-transformed market values of single-family residential properties in the City of252Cincinnati; S is a matrix of structural characteristics of the properties, C is a matrix of quadratic controls of253longitude and latitude, and N is a matrix of neighborhood characteristics, including the distances from each254property to the closest park (parkdist), the sizes of the closest park (parksize), and the presence of park255facilities (parkfacility). The four vectors β, γ, θ, and ϕ are the corresponding parameters to be estimated256and W is the spatial weight matrix. The fact that economic theory does not imply the functional form of257hedonic models (Bolitzer and Netusil, 2000) and the common practice of using log-transformed prices, while258not transforming distance measures (Czembrowski and Kronenberg, 2016), we chose a log-linear functional25973260261262===== PAGE 11 (score 29, numeric-paren 0, table-words 4, var-words 8) =====263264Journal of Regional Analysis & Policy26548(3): 66–82266Table 3: Effects of Parks on Residential Property Values in the City of Cincinnati (n = 36,167)267Variables268SDEM269SLX270Parameter271p-Value272Parameter273p-Value274constant27510.201465276<0.0127710.211803278<0.01279Structural characteristics of the properties280sizeland2810.000053282<0.012830.000048284<0.01285sizehouse2860.002684287<0.012880.002859289<0.01290bathrooms2910.051512292<0.012930.055278294<0.01295age296-0.010605297<0.01298-0.010174299<0.01300agesquared3010.000038302<0.013030.000036304<0.01305garage3060.027471307<0.013080.017696309<0.01310condvpoor311-0.923445312<0.01313-1.072054314<0.01315condpoor316-0.49363317<0.01318-0.514455319<0.01320condfair321-0.219315322<0.01323-0.211597324<0.01325condgood3260.135936327<0.013280.124734329<0.01330condvgood3310.291110332<0.013330.262003334<0.01335condexcellent3360.441086337<0.013380.465527339<0.01340Neighborhood characteristics341poverty342-0.001252343<0.01344-0.001140345<0.01346bachelor3470.011986348<0.013490.011502350<0.01351crime352-0.001909353<0.01354-0.001070355<0.01356school3570.001646358<0.013590.002347360<0.01361parkdist362-0.0000513630.018364-0.000028365<0.01366parksize3670.000183368<0.013690.000181370<0.01371parkfacility372-0.037543373<0.01374-0.043141375<0.01376Structural characteristics of neighboring properties377W ∗sizeland3780.000001379<0.013800.0000033810.135382W ∗sqmtrfinished3830.000833384<0.013850.000770386<0.01387W ∗bathrooms3880.032671389<0.013900.026945391<0.01392W ∗age393-0.006336394<0.01395-0.005147396<0.01397W ∗agesquared3980.000029399<0.014000.000024401<0.01402W ∗garage4030.038363404<0.014050.044361406<0.01407W ∗condvpoor408-0.489707409<0.01410-0.669583411<0.01412W ∗condpoor413-0.256236414<0.01415-0.282074416<0.01417W ∗condfair418-0.208881419<0.01420-0.230919421<0.01422W ∗condgood4230.145953424<0.014250.101972426<0.01427W ∗condvgood4280.307800429<0.014300.237923431<0.01432W ∗condexcellent4330.3704854340.034350.435356436<0.01437λ4380.387626439<0.01440—441—442R24430.7714440.6657445AkaikeInformationCriterion(AIC)44661,51644763,69744876449450451===== PAGE 12 (score 16, numeric-paren 0, table-words 0, var-words 15) =====452453Journal of Regional Analysis & Policy45448(3): 66–82455is worth $14,937 more than the house in average condition. Houses in worse than average condition lose456$5,306, $10,497 or $16,242 when their condition drops to fair, poor, or very poor respectively.457Neighborhood-related characteristics show that the poverty rate and the crime rate have, as expected, a458negative impact on the sales price of a property; for every percent of the population below the poverty rate the459price decreases by 0.13% ($154) and for every extra crime per 1000-population, by 0.19% ($235). Meanwhile,460every 1% of people with bachelor’s degree in the census tract increases the sales value of a property by 1.20%461($1,473) and every 1 point increase in the school performance indicator adds 0.16% ($202) to the market462value.463Turning now to the variables of interest, the characteristics of the park located closest to each property,464we find as expected that the price of a property decreases as the distance from the closest park increases.465For our average property valued at $122,852 this means that for every 100 meter increase in the distance466to the closest park, the sales price of a property decreases by 0.51% ($627), or $6.27 per meter change in467distance. As such, our finding confirms research by Correll et al. (1978), Crompton (2001); Lutzenhiser468and Netusil (2001), among others, who find increases in the value of properties surrounding parks in similar469magnitude using buffer zones from approximately 150 to 1,000 meter. The size of the closest park, although470significant, has only a minimal influence on property prices. For our average house of $122,852, increasing471the size of its closest park by 1 hectare would merely increase its value by 0.018%, or $22.48. Completely472different is the outcome regarding recreational uses of the parks. If the closest park includes a recreational473facility, the average decrease in the price of a property is of 3.75%, or $4,612. Our results confirm earlier474studies by Crompton (2004); Espey and Owusu-Edusei (2001); Lutzenhiser and Netusil (2001); Shultz and475King (2001) who showed that parks providing more natural landscapes and passive uses tend to have a476greater, positive impact on surrounding residential values. On the other hand, we only find a relatively477small incremental value of slightly larger parks versus smaller parks, considering that our average house’s478closest park is 45.1 ha. It also leaves open the question for further research, whether the detrimental value479of parks with recreational facilities stems from the pure presence of the recreational facility, or is attributed480to its often poor condition and neglected maintenance. This issue also raises the question of what types of481recreational facilities are preferred over others.482The spatial spillover effects in the SDEM model483captured by the θ coefficients in equation (1)484are485purely of local nature (LeSage and Pace, 2014). As such, the θ coefficients capture how residential property486prices are influenced by changes in structural characteristics of immediately neighboring properties. All local487spillover in the SDEM model, with the exception of the W ∗sizeland and W ∗bedrooms coefficients, are488statistically significant. Adding 1 square meter to the average house size of the adjacent properties induces an489increase by 0.08% ($102) in the market value. The other structural characteristics, averaged over neighboring490properties, also induce similar, though lower, effects as compared to own structural characteristics (3.27%491($4,014) for 1 extra bath, minus 0.63% ($778) for 1 year of age and 0.0029% for agesquared, and 3.84%492($4,713) for 1 more space in the averaged capacity of the basement garage). Also interesting is the relatively493large impact the neighboring houses have on the average house with respect to the conditions they are494in. Very poorly- or poorly maintained neighboring houses negatively affect own market values by $10,432495and $6,090 respectively, while properties in fair condition still depreciate own market values by as much496as $5,079. Contrarily, neighboring houses that are in above average condition positively affect own market497values by $4,234, $9,711, and $12,082 when they are in good, very good, and excellent condition. Due to498potential multicollinearity issues, we excluded all neighborhood variables–including all three park-related499characteristics and distance to the CBD–from the construction of the spatial lags in the SDEM.500When selecting the most appropriate model framework, we ended up, for reasons discussed above, with501two potential spatial models: the SDEM model and the SLX model. Though the SDEM model provides502the overall better fit to our data sample, the SLX model also provides some interesting findings. Comparing503our direct effects from the SDEM and the SLX models confirms our general conclusions. The estimated504coefficients for the SLX model all have the expected signs and are all significant, except for the lagged size of505the land parameter (p = 0.135). A direct comparison of the SDEM and SLX model coefficients reveals that506the majority of the estimated coefficients are similar in magnitude, but exhibit, on average, differences of5079.7% for the structural characteristics, 22.9% for the neighborhood characteristics, and 18.6% for the lagged508structural characteristics. Examining the three estimated parks parameters in the SLX specification–park509distance, park size, and park facility–we find that for every 100 meters increase in the distance of our property51077511512513===== PAGE 13 (score 61, numeric-paren 44, table-words 4, var-words 3) =====514515Journal of Regional Analysis & Policy51648(3): 66–82517valued at $122,852 its market value decreases by 0.28% ($344), or $3.44 per meter. A change in the nearest518park’s size by 1 hectare translates into a 0.02% ($22.24) increase in value of our average property. The519interesting finding, however, is that according to the SLX result, recreational facilities decrease the market520value of the nearest residential property by 4.31% ($5,300) which again emphasizes that residents prefer521passive park uses over active ones.522One of the important choices we had to make when assessing the influences parks have on residential523property values is whether or not to use a buffer zone around the parks for identifying the properties to be524included in our study. Given that the relevant literature is inconclusive as to whether to apply buffer zones,525or as to the size of these buffer zones, we conducted a sensitivity analysis to find out how the parks impacts526differ for different buffer zones in our study area. Our findings from this sensitivity analysis are presented527in Table 4.528Table 4: Sensitivity of Park Effects on Property Values with Respect to Size of Buffer Zones529Buffer530Spatial Durbin Error Model531Number of532Zones533parkdist534parksize535parkfacility536Observations (n)5370-100 m5380.0000225390.000289540***541-0.029609542**5436,3405440-250 m545-0.0000715460.000249547***548-0.035509549***55018,415551250-500 m552-0.0000435530.000126554***555-0.037637556***55717,752558100-500 m559-0.0000355600.000167561***562-0.037449563***56429,8275650-500 m566-0.000051567**5680.000183569***570-0.037543571***57236,167573500-750 m5740.0000155750.000147576***577-0.038445578***5799,145580750-1000 m5810.000346582***5830.000053584-0.033261585*5863,989587500-1000 m5880.000229589***5900.000146591***592-0.037838593***59413,134595no buffer5960.000061597***5980.000141599***600-0.034785601***60251,786603Buffer604Spatial Lag of X Model605Number of606Zones607parkdist608parksize609parkfacility610Observations (n)6110-100 m612-0.0001316130.000335614***615-0.044068616***6176,3406180-250 m619-0.000063620**6210.000274622***623-0.041457624***62518,415626250-500 m627-0.0000156280.000119629***630-0.044139631***63217,752633100-500 m634-0.0000096350.000152636***637-0.044089638***63929,8276400-500 m641-0.000028642***6430.000181644***645-0.043141646***64736,167648500-750 m6490.0003706500.000091651***652-0.035325653***6549,145655750-1000 m6560.000319657***6580.000063659**660-0.026550661***6623,989663500-1000 m6640.000188665***6660.000106667***668-0.035218669***67013,134671no buffer6720.000048673***6740.000122675***676-0.037216677***67851,786679* significant at the 0.10 level. ** significant at the 0.05 level. *** significant at the 0.01 level.680Not surprising, changing the distances for the buffer zones significantly affects the outcomes of the681estimated parameters. First, the park distance parameter (γparkdist) switches sign with increasing distances,682i.e., after 500 meters, in both the SDEM and SLX model specifications. While market values within 500683meters increase as distances to the nearest park decreases, outside the inner 500-meter buffer the results684indicate that market values tend to increase with increasing distances. Unfortunately, the fact that half of685the estimated park distance parameters (ˆγparkdist) are not significant does not allow valid conclusions to686be drawn with respect to their changes in magnitude. Nevertheless, the results allow a comparison of the687first 0-500 meter buffer to the second 500-1000 meter buffer. Here we find a park effect for the 0-500 meter688buffer of negative $6.27 (ˆγparkdist = −0.000051) per meter increase in distance from the park compared to a689positive $28.13 (ˆγparkdist = 0.000229) per meter in the second 500-1000 meter buffer. Both of these values690are from the SDEM model in reference to our averagely priced property of $122,852. This finding indicates691not only the sensitivity of the nearest park effect on property values, but also questions the park effects on692property values beyond 500 meters in our study area, or beyond what we identified as a walkable distance693to the nearest park.69478695696697===== PAGE 14 (score 6, numeric-paren 0, table-words 0, var-words 6) =====698699Journal of Regional Analysis & Policy70048(3): 66–82701Second, park sizes matter regardless of how far a residential property is from the nearest park. In both702the SDEM and SLX spatial model specifications (with significance of all but one estimated parameter),703larger park size matters regardless of the distance to the nearest park. As expected, we find that property704values generally tend to increase with an increase in park size and that the positive effect is stronger for705properties closer to their nearest park. Third, park facilities do have a negative effect on property prices, as706indicated by the significance of all estimated parameters in both our spatial models. For the SDEM model,707the negative effect of park facilities appears to be fairly constant with respect to applied buffer zones, except708the 0-100 meter zone (-0.029609), which is somewhat smaller in magnitude. For the SLX model, the effect of709park facilities is definitively larger within the first 500 meters, but decreases significantly in size beyond the710500 meter threshold. To complete this sensitivity analysis of how the park facilities differ in impact across711the various buffer zones in our study area, we included the estimated parameters when not using buffer zones712at all. Here we find that the park facility effects are similar to the effects when applying buffer zones of713greater than 500 meter. Again, the park distance parameters, while significant, do not have the expected714sign.7155716Conclusion717It is a well-known fact that location matters for homebuyers and homeowners alike, and plays an important718role when determining property values. In this paper, we show that urban parks have a significant influence719on house prices when they are located within close proximity to a park, where the distance to the nearest720park is calculated based on Euclidean distances. More specifically, using the results from a spatial hedonic721pricing model that accounts for local spatial dependence among residential property prices–the Spatial Durbin722Error Model (SDEM)–we evaluated three different park influences on property values. First and foremost,723we find that proximity of residential properties to parks benefits property owners. The price of a property724increases with decreasing distance to the closest park, which confirms previous findings. More specifically,725we estimated the influence of the Cincinnati parks in Hamilton County, Ohio, to increase the average priced726house of $122,852 in our sample by $6.27 when moving closer to the park by one meter. Secondly, increasing727the size of a nearby park by one hectare increases the average priced house by 0.018%, or $22.48. While this728confirms other research that park size maters, the magnitude of the park size influence on property values729is of lesser importance in Cincinnati, which one might explain by the large number of parks within the city.730Last, we find that properties located close to parks with recreational facilities can register a total decrease in731the price of a property of 3.75% ($4,612) for the average property according to our results. This result raises732the issue of whether the detrimental value of parks with recreational facilities stems from the pure presence733of the recreational facility, or can be attributed to its often poor condition and neglected maintenance. It734also leaves open the question of what types of recreational facilities are preferred over others.735In this paper, we presented a sensitivity analysis with the estimation results when using different buffer736zones. Overall, we conclude that the estimated results for the three park parameters–park distance, park737size, and park facility–are very sensitive with respect to the chosen buffer zone. Most interesting is the738finding that with increasing distance to the nearest park, property values clearly decrease for houses within739500 meters to the park. Beyond this 500 meter buffer, we find exactly the opposite effect; with increasing740distance, houses appear to appreciate in value. We also find that the bigger the park, the better. The size741of the nearest park positively influences property values, regardless of the distance of the property to the742nearest park. Park facilities, on the other hand, tend to negatively affect property values, if properties lie743within a 500 meter distance to the nearest park. This negative influence does, however, not extend beyond744the 500 meter, as indicated by our sensitivity analysis.745Overall, the findings of this sensitivity analysis746indicate that more research is needed to more fully understand the influence on the estimated parameters747stemming from various buffer zones. In addition, individual park facilities may also have different effects on748property values, a fact that suggests further research of this matter.74979750