Hierarchical Clustering Methods With Territorial Integrity Criterion

Abstract

During the conference entitled Spatial Econometrics and Regional Economic Analyses, which took place in Lodz in 2014, there was a proposition to introduce the spatial coherence property into the Ward method, which is applied to group administrative units [1]. At each stage of agglomeration in the modified Ward method, there are included only those aggregates which are adjacent to each other. This work is an extension of this concept based upon other methods of hierarchical clustering, in particular the single and complete linkage method. The study highlighted the benefits of cluster­ing methods with the coherence property, also emphasizing the limitations of these procedures. First of all, the introduction of the restricting condition during the procedure of the hierarchical clustering reduces the homogeneity of isolated clusters. Spatial constraints may also lead to a situation where the distance between the clusters linked at a later stage is smaller than at an earlier stage (graphically, we can talk about the dendrogram “backflow”). The method of complete linkage is free of these aberration where the distance between the clusters is defined as the maximum distance between their elements. The modified clustering algorithm was implemented as an extension of STATISTICA software. Examples of an application of the hierarchical clustering method with the coherence concern sector changes in the European regional space. The aim of the analysis was to isolate spatially coherent areas that demonstrate a similar direction and intensity of structural change in selected areas of the labour market.