Information Aggregation in Probabilistic Prediction

Probabilistic prediction generally involves the consideration of information from many different sources, and this information must be aggregated to determine a single probability (or probability distribution). This paper is concerned with the aggregation process, and although some aspects of the paper are new, much of the paper is tutorial in nature. Models of the aggregation process are discussed, with particular emphasis on the question of the conditional dependence of information, and measures of the redundancy of information are developed. In addition, a review of previous experiments concerning the aggregation process is given, along with suggestions for experiments that should provide additional insight into the nature and ``efficiency'' of this process. In view of the importance of probabilistic prediction in inferential and decision-making situations, additional investigation and experimentation concerning the aggregation process should be of considerable value.