Bayes Factors from Mixed Probabilities
混合概率下的贝叶斯因子
Summary A new expression for a Bayes factor as a weighted average of conditional Bayes factors is given, resembling the familiar relation for a marginal probability as a weighted average of conditional probabilities. Interpretations are given for the three contexts of (i) the use of a non-sufficient statistic, (ii) inference for a simple hypothesis versus a vague alternative in terms of inference for a dichotomy and (iii) opinion pooling before sampling.