Non symmetrical data analysis via statistical implication

For many methods of data analysis, the data are organised according to a criterion of similarity measured by some index, which is usually a symmetrical one. However, in many real situations, it is necessary to structure a set of variables or variable classes according to inclusion or inferential relationships, such as ‘If A, then B’. The approach in this paper (inspired by the works of I. C. Lerman) is to find an orientated classification of such variables, visualised as a graph. This yields a hierarchy of variables or sets of variables. The paper goes on to examine the significant nodes of the graph as well as the contribution of subsets to the overall structure of the data.