Résumé
We aim at characterizing graphical models for convolution splitting distributions. Only marginal independence have been studied through the well-known Rao–Rubin condition. We generalize this condition for conditional independence and deduce the desired characterizations.