Hierarchical explanation of inference in Bayesian networks that represent a population of independent agents