On the modeling of causal belief networks