The role of cue information in the outcome-density effect: Evidence from neural network simulations and a causal learning experiment