It's a question of using the right design for the right problem. Does it make sense conceptually for an object of type A to contain a collection of other objects of the same type? That will depend on what A actually represents, and what relationships between objects you want to model.