贝叶斯预测集成用于随机效应不确定性的建模

A common challenge in multilevel modeling is determining which predictors should have random effects, especially in educational and psychological research where several theoretically plausible model specifications exist. Choosing a single specification ignores information from the alternatives and can lead to understated uncertainty in later inference. Popular Bayesian ensemble methods, including Bayesian model averaging (BMA), Bayesian stacking (BS), and Bayesian hierarchical stacking (BHS), combine predictions across candidate specifications but constrain model weights to sum to one, implici