Likelihood judgment based on previously observed outcomes: the alternative-outcomes effect in a learning paradigm