The Myth of Reduction : Or Why Probabilities Average over Counterfactual Hidden Variables