Un nuovo indice di associazione: applicazione al paradosso di Simpson

Abstract

In the study of association between two dicotomic variables A and B, in presence of a third variable C, it is possible to derive some relations involving the relative frequencies of data. These relations, when we put together data ignoring the third variable C , permit to avoid changes in the sign of the index measuring this association (Simpson Paradox). They have been analysed in detail by geometric approach which provides also an evaluation of the probability (depending on the above frequencies) that the paradox occurs.