You can show that the likelihood (read: P(data|~X) of ~X is low, given a particular underlying model and a set of observations. To bootstrap that to a probabili

You can show that the likelihood (read: P(data|~X) of ~X is low, given a particular underlying model and a set of observations. To bootstrap that to a probability of ~X (read: P(~X | data) ), you need philosophy to help choose a prior on the underlying set of models. Or of course you can just ignore these details and fallaciously conflate likelihood and probability like most pop-frequentists, but that leads to crazy results even in the "real world."