The classification of objects, variables, constructs in proper groups is very important for theory development. Traditionally, numerical clustering techniques such as cluster analysis and discriminant analysis have been used for... more
The classification of objects, variables, constructs in proper groups is very important for theory development. Traditionally, numerical clustering techniques such as cluster analysis and discriminant analysis have been used for classifying objects into similar and dissimilar groups. These techniques use a numerical measure of similarity or dissimilarity to cluster objects. Such a measure is a function only of the compared objects and does not take into consideration any global properties or concepts characterizing object classes. Consequently, the obtained clusters may not be simple and may be difficult to interpret. A recently developed conceptual clustering technique (Michalsky, 1980) which forms clusters only if they are describable by a concept from a predefined concept class has been claimed to give better clusters than using conventional clustering techniques (Michalsky and Stepp, 1983, 1986). However, the results obtained from different studies so far are contradictory and inconclusive. In this study, the above claim is examined by making a comparison between conceptual clustering and conventional numerical clustering on the basis of simplicity of the generated clusters and the simplicity of concept description for the interpretation of the generated clusters. The study is grounded in past research and suggests the superiority of the conceptual clustering over conventional numerical clustering. However, the extensive empirical testing is necessary to substantiate the above claims. The limitations of the present conceptual clustering approach are identified and the future research directions are suggested.