On comparison of non-Bayesian experts

An expert, trying to assess the true distribution over the states of nature, is associated with a preference relation over utility bundles. He prefers f to g if he believes that, according to the true distribution, the expected utility of f is greater than that of g. Expert I is said to be more knowledgeable than expert II (Lefort, 2009) if, between the two experts, it is always beneficial to follow his advice. It is shown that if experts' maxmin preferences are induced by information consistent with the true distribution, then in most cases an expert having more information is not more knowledgeable.