An investigation of the effects of multiple distance measures in regression modeling: Predicting house prices

Publication Year:
Usage 30
Downloads 16
Abstract Views 14
Repository URL:;
Porter, Dawn Maire
Human Geography
thesis / dissertation description
In today’s society people’s locational residential preference is no longer dependent on the distance from the Central Business District. With the demise of the concept of the friction of distance other accessibility nodes, besides the central business district, such as retail and suburban employment are necessary attributes towards the determination and influence of land rents. Where this is the case, the hedonic regression methods analysis to explain house prices should employ distance variables corresponding to each of the urban nodes. However, these distance measures may be highly intercorrelated, thereby posing a problem of “spatial collinearity.” Two authors have examined and attempted to reduce spatial collinearity from a purely contrived theoretical level. They found that problems arising from spatial collinearity can be avoided or substantially lessened by carefully selecting the geographic domain from which observations are drawn. This thesis explores the problems of spatial collinearity from an empirical perspective through the use of a data set containing housing characteristics and prices from the Kitchener-Waterloo area. A regression model was run, such that the calculated spatial collinearity could be compared to the theoretical findings from one of the above authors. It was found that the extent of collinearity is indeed influenced by the spatial configuration of nodes relative to the data range; however, a discrepancy occurred in the type of pattern that was needed to reduce this. Instead what was discovered was that the optimum spatial configuration of the data was not as relevant to the reduction of spatial collinearity as was the actual distances between the nodes.