Defining apartment neighbourhoods with Thiessen polygons and fuzzy equality clustering

The purpose of the paper is to verify whether the version of neighbourhoods created from the lowest geographical level improve a predictive accuracy of hedonic model in comparison with those based on upper geographical levels. Methodology/approach ñ The paper proposes a method for defining neighbourhoods from Thiessen polygons created around the points of apartments. These polygons occupy the whole analysed area and are used as the spatial units for clustering. The clustering technique is based on contiguity of polygons and fuzzy equality of the principal components of their attributes. Clustering is started at different geographical levels: municipalities, smaller traffic analysis zones, and apartmentsí Thiessen polygons. The ordinary least squares (OLS) and spatial error techniques are applied in hedonic price models with different versions of neighbourhoods. Originality/value ñ Neighbourhoods can be defined using the Thiessen polygons of individual observations. This very ìbottom upî approach can minimise dependency from existing political, administrative and other boundaries. The clustering technique is based on fuzzy equality and does not need the a priori determination of a number of clusters, while contiguity and hierarchical nature of neighbourhoods are considered. Findings ñ With OLS regression, the superiority of Thiessen polygons is evident in both in-sample analysis and ex-sample prediction. When we control for spatial effect with a spatial error technique, the clusters of Thiessen polygons do not always provide the best outcome, and their superiority is contested by the highest geographical level of municipalities.