arXiv Study: Matching Probability Combination Rules to Situations and Stakes

A recent study published on arXiv (ID: 2608.11275) explores the integration of numerical evaluations for binary questions, highlighting that the correct formula is contingent on the meaning of the numbers and the interrelation of the sources. The research outlines the assumptions underlying various common combination methods and calculates the associated combined probabilities. It asserts that averaging is valid under specific alternative interpretations, while multiplying odds is appropriate when probability reports stem from conditionally independent evidence and a shared prior. Additionally, it emphasizes that the probabilities of different successful outcomes must consider dependence or common evidence. Two distinct groups of Monte Carlo experiments investigate different aspects: the first confirms that the derived rule accurately retrieves the correct probability under its assumptions, and the second assesses the impact of employing an incorrect rule through logarithmic scoring. This paper appears to be a cross-type submission on arXiv, suggesting it may be relevant to multiple categories. The authors remain unnamed in the text. This work enhances decision theory and probability aggregation, with implications for AI, forecasting, and risk assessment. The full paper can be accessed at https://arxiv.org/abs/2608.11275.