When the Unreal is More Likely Than the Real: Post Hoc Probability Judgements and Counterfactual Closeness