The bias of the results of cluster analysis for ordinal data using distances for metric scales

<p>&nbsp;</p><p>This paper is related to methodological scope of sociology. It concerns a problem of realization of cluster analysis based on ordinal variables. Certainly, there are two ways of performing such statistical operation on ordinal scales. However in both cases the researcher has to modify variables. The main research issue is to figure out is there a bias of cluster analysis results when the researcher uses metric scales&rsquo; distances.</p><p>This work contains three chapters: the first one is a kind of literature overview built on several theoretical and methodological works, which are relevant to the subject of this research. Also the first part of this paper performs a detailed consideration of cluster analysis. The second chapter contains the experiment description and the third one is devoted to results of experiment.</p><p>Because of a lack of similar researches in sociology it is possible to say that this work has experimental character. The algorithm of experiment consists of several main stages. Firstly, there is a clustering of objects in three different ways and one of this ways uses Kemeni-Snell&rsquo;s distance suitable for multidimensional classification, but almost not being used in practice. And secondly, there is a comparison of the received results.</p><p>For ensuring the clearest comparison the artificial database with the existing ideal solution of clustering problem is in a basis of experiment.</p><p>During the carried-out data analysis our general hypothesis about the possibility of bias existence was confirmed. We consider it means that this scope of sociology needs some more similar investigations. And it also illustrates the usefulness of this paper for sociology.</p>