Conditionalization and observation (Q1393902): Difference between revisions

This paper develops arguments for the Bayesian rule of conditionalization. It sketches an argument for conditionalization from a frequentist point of view, and then presents in detail a Dutch book argument for conditionalization devised by David Lewis. The bulk of the paper develops a method of studying conditionalization in terms of qualitative considerations. First the paper proves that under suitable background assumptions a change from one probability measure to a second takes place if and only if a certain qualitative and, psychologically speaking, very simple relation holds between the two measures. Second, this equivalence is used to suggest a plausible qualitative principle of inductive reasoning, such that if two systems of beliefs are characterizable by probability measures and if the conditions of application of the principle hold, the second system of beliefs cannot be reasonable unless it arises by conditionalization from the first. Third, the paper further specifies the conditions of application of the principle in terms of a partial characterization of observational knowledge. Finally, the paper develops all of this material for Richard Jeffrey's generalization of conditionalization, which does not require certainty or any probabilities of one or zero.