Seminario 17/01: – Jose G. Clavel (U. Murcia)- Total Information Analysis: Representation of Categorical Data Through Clusters

Información

Abstract:

Due to the presence of duality, cluster analysis has been proposed to analyze Total Information Analysis super distance matrix instead of a row-column joint graph. But the same duality causes that the traditional methods of clustering are not always capable to produce meaningful results. In this paper following Nishisato (2014), the clustering with p-percentile lters is applied to several cases and the results are compared with the traditional methods. The analysis using traditional methods |in particular hierarchical and partition approaches are not able to discover the relationships between rows and columns when the distances within elements are smaller than the distances between them. In theses cases, the comprehensive analysis of the relationship between
rows and columns can only be made clustering with p-percentile lter.