Abstract
This study was conducted using the arms-length residential property transactions in Onondaga County, NY, from January 2000 to December 2000. I developed a hedonic model that investigates how different types of urban open space and their spatial configurations are valued by residential property owners not only with traditional hedonic variables explaining the structure and neighborhood characteristics of residential properties but also with environmental amenity variables describing environmental characteristics of open space surrounding each residential property. Specifically, four spatial metrics from landscape ecology are included in the hedonic framework to quantify the spatial configuration of urban open space on residential property values. In order to capture scale issues in the analysis, environmental amenity variables are measured at 100m and 1km radius buffer surrounding each residential property. The first scale described the visual zone around a house and the latter scale was more representative of a typical walking distance. To select the important independent variables for hedonic models for both sizes of buffers, machine learning approaches including Cubist and random forest were applied and compared with standard Ordinary Least Squares (OLS) regression method. The results showed machine learning approaches outperformed standard OLS regression method in terms of model fitting and model accuracy. To deal with spatial heterogeneity and spatial autocorrelation issues which are common in hedonic study, Geographically Weighted Regression (GWR) method was applied and compared with OLS regression. By integrating GWR and spatial metrics into the hedonic framework, I investigated how urban residents value the spatial configuration of their neighborhood. The results found that house buyers positively value environmental amenities from various types open space and its spatial configuration at the same time. In addition, within the smaller buffer, property values were higher when land uses were more homogenous but within the larger buffer, house buyers prefer more heterogeneous landscape. Moreover, the results showed that the application of GWR reduces these spatial effects in two different buffers.