On Causal Explanations in Bayesian Networks