O'Hagan [38] gives a somewhat contrived example of why it is important to consider the prior as well as the likelihood. Let G be the event of seeing a big green structure, with blob like attachments outside a window. Let T be the hypothesis that a tree is outside the window, and let C be the hypothesis that a cardboard model is outside the window. Since C and T are equally consistent with the observation, G, one shouldn't have any reason for believing one over the other. That is l(C|G) = l(T|G). However, the probability that C is in fact outside the window, conditional on the observation, is p(C|G), which depends on p(C), the prior probability of cardboard structures being outside windows, and is likely to be much less than p(T|G). Incorporation of prior knowledge is an essential part of the inference.