On the validity of long-run estimation methods for discrete-event systems