Clustering Using Multidimensional Indicators: An Approach without Feature

The complexity of social phenomena can be decomposed into different dimensions, each measured by a set of basic indicators. These elementary measures capture distinct aspects of each dimension, summarizing them into a composite indicator. In this framework, one of the main goals of statistical analysis is to partition individuals using these dimensional indicators directly rather than all available sets of indicators, possibly leading to a loss of information. In this work, we propose a new methodological approach that allows the clustering of individuals using all the elementary measures, preserving information about the dimensional structure of the phenomenon under study when characterizing the clusters obtained from the partitioning algorithm. Our method relies on distance matrices between individuals belonging to different groups with elements that possess the desirable property of multivariate additivity. The results of applying this method to an example dataset are compared with those obtained using traditional methods.