Using Meta-Analysis Results in Bayesian Updating: The Empty-Cell Problem
指出在贝叶斯估计中，使用元分析结果作为先验信息时，由于元分析设计中存在大量空单元格，会导致先验有偏，并发现设计简化是产生无偏先验的优选方案。
Bayesian estimation incorporating prior information has been a popular approach to gaining estimation efficiency. Although prior information can take a variety of forms, generalizations derived from meta-analyses have been suggested as being useful. This article shows that these priors can be problematic in light of the many empty cells observed in meta-analysis designs. Design reduction, which gives rise to an unbiased prior, is found to be the preferred solution.