Finitely Additive Priors, Coherence and the Marginalization Paradox
William D. Sudderth · Journal of the Royal Statistical Society Series B (Statistical Methodology) · 1980
Summary The marginalization paradox, which arises in a Bayesian theory of inference based on improper, countable additive prior distributions, is resolved when proper, finitely additive distributions are used instead. The resolution is related to the equivalence of coherent inferences to inferences from proper, finitely additive priors under a certain definition of “coherence”.