Probabilistic Opinion Pooling

This review article introduces and evaluates various ways to aggregate probabilistic opinions of different individuals. For each of these three ways, an axiomatic characterization result is presented (a new one in the case of multiplicative pooling). The three ways satisfy different axioms and are justifiable under different conditions. Linear pooling may be justified on procedural grounds, but not on epistemic grounds. Geometric and multiplicative pooling may be justified on epistemic grounds, but which of the two is appropriate depends not just on the opinion profiles to be aggregated but also on the information on which they are based. Geometric pooling can be justified if all individuals' opinions are based on the same information, while multiplicative pooling can be justified if every individual's opinions are based solely on private information, except for some shared background information held by everyone.