Contrasting cue-density effects in causal and prediction judgments