How exactly does one get a posterior probability of signal+background > 50% from this data-set? You’d need some pretty tight priors on the location & width of the signal peak to get those sorts of odds. And from scale of model fitting currently going on, I’d say that was highly unlikely.

]]>Everyone in the field obviously knows that the “local” p-values vastly overstate the probability that the signal is real. In spite of that, the “actual reasonable” probability that the signal is real may very well be above 50% now.

]]>