Niehierarchiczna analiza skupień — nowa metoda klasyfikacji zjawisk społeczno-gospodarczych

Abstract

Two manners of class separation in a unit set are known: 1) set division into homogeneous parts, and 2) grouping similar elements. Nonhierarchical cluster analysis presented in the article is a classification method based on unit grouping It can be widely used and there is no need to accept preliminary assumptions on structure of examined set of elements. It is, however, in the case of hierarchical grouping techniques. The discussed mthod can be used not only as a method of hypotheses testing but above all as a methods of their creation. Following steps exist when non-hierarchical cluster analysis is used: 1) determination of an approximate number, of classes (a simplified picture of distribution of points representing set elements in two-dimensional space is used to this end), 2) identification of cluster cores by: a) arbitrary choice of typical units from observations matrix, b) random choice of cores from values between the minimum and the maximum of each feature observed, 3) distribution of units to cores on the ground of calculated distance, and 4) statistical appraisal of solution correctness (f.ex. F Test). Each step has been discussed in the first part of the article in detail. An example of utilization of non-hierarchical cluster analysis for determination of types of socio-economic structure of the Wielkopolska villages was presented in the second part of the elaboration. The obtained results were compared with the country typology determined on the ground of the Czekanowski's diagram. It allowed to present advantages and imperfections of non-hierarchical cluster analysis.