unpaved recreational greenway trails. Lakes and streams were identified using GIS datasets available from the Minnesota Department of Natural Resources.
Distances were calculated within a GIS in two ways. First, we calculated Euclidean distances between the centroid of each subject residential property and the closest trail, lake, and stream. We then calculated road distances between each property and the closest large park. We calculated road distances to parks because we felt that they would better approximate homeowners perceptions of their access to open space as they would likely access these areas on roads rather than by cutting across other parcels in an as-the-crow- flies fashion. Initial model runs indicated that this was indeed the case as none showed significant results for the impact of Euclidean distance to parks on home sales prices. Thus, four open space access variables were used in the final analysis: Euclidean distances to the closest (1) trail, (2) lake, and (3) stream, and road distances to the closest (4) park.
To provide a measure of the scenic quality of the landscape surrounding each residential property, we calculated viewsheds using the VIEWSHED function in ArcGIS. This function computes the locations in a DEM that are connected to an observation location, in this case, each home sold in 2005, by a line-of-sight within a specified distance accounting for the location, height, and angle of view of the viewer in three dimensions. Because existing DEMs for the area did not include aspects of the built environment likely to obstruct views, we constructed a DEM for use in viewshed calculation using the following steps. First, we obtained a GIS dataset containing the footprints and locations of all Ramsey County buildings over 7.5 m2 in area from the Ramsey County Surveyor’s Office. We assigned a height to each building based on its land use type, if a parcel’s land use classification was not residential, or based on its dwelling type, if a parcel was classified as residential, as indicated by the parcel dataset. The mean number of stories in buildings of each type was identified via visual surveys of 20 buildings of each type and multiplied by three, assuming stories to be 3 m on average, and