Robustness of Evidential Probability

In the Evidential Probability (EP) (Kyburg and Teng 2001) framework, interval-valued probabilities are assigned to events given a knowledge base containing statements about logical relationships of classes of objects as well as statements about relative frequencies pertaining to some of those classes. We are interested in the robustness of this inference with respect to perturbations in the given knowledge base. First we define how changes to a (non-deductively-closed) knowledge base are to be carried out by constructing revision and contraction operators that satisfy the AGM postulates, and then we discuss some conditions and measures of robustness. We appeal to a result in (Kyburg and Teng 2002) that allows a logic of risky knowledge based on EP to be characterized by the classical modal system EMN, and define AGM operators for EP through a mapping from monotone neighborhood modal structures to polymodal Kripke structures that admit a first-order correspondent.