A first inquiry into Simpson's paradox with belief functions
Résumé
Simpson's paradox, also known as the Yule-Simpson effect, is a statistical paradox which plays a major role in causality modelling and decision making. It may appear when marginalizing data: an effect can be positive for all subgroups of a population, but it can be negative when aggregating all the subgroups. This paper explores what happens if data are considered in the framework of belief functions instead of classical probability theory. In particular, the co-occurrence of the paradox with both the probabilistic approach and our belief function approach is studied.
Origine : Fichiers produits par l'(les) auteur(s)