基于贝叶斯模型平均的元分析：原理与实现
Meta-Analysis Based on Bayesian Model Averaging: Principles and Implementation
摘要
元分析作为一种综合已有研究成果的重要统计方法，在量化研究中广泛应用。然而，研究者在分析过程中常面临模型选择的困境：在处理研究间异质性时，需在固定效应与随机效应模型之间做出选择；在应对潜在发表偏倚时，又需要从多种校正模型中选定一种。目前，这些模型选择仍缺乏统一标准。贝叶斯统计框架下的贝叶斯模型平均(Bayesian Model Averaging, BMA)为解决这些问题提供了新的思路：它将不同统计模型纳入同一模型空间，量化各模型获得数据支持的程度，并据此对效应量进行加权，从而规避单一模型选择带来的不确定性。基于BMA的元分析能同时检验三个关键假设(效应是否存在、异质性是否存在以及发表偏倚是否存在)，并以模型平均的方式实现对效应量的稳健估计。该方法可通过开源软件 JASP 或 R 语言实现，为研究者实施元分析提供了新选择。
Abstract
Meta-analysis is widely used to synthesize independent findings, yet researchers often face difficult choices among fixed- and random-effects models and multiple publication-bias corrections. Conventional analyses typically select a single model, ignoring model uncertainty and potentially producing overconfident inferences, underestimated standard errors, and biased effect estimates. Bayesian model averaging (BMA) addresses this limitation by treating models as random variables, assigning prior probabilities to plausible candidates, updating them with data, and weighting model-specific estimates by posterior model probabilities. Robust Bayesian meta-analysis (RoBMA) extends BMA by jointly considering the presence of an overall effect, between-study heterogeneity, and publication bias. It incorporates selection models and PET-PEESE corrections within a unified model space and evaluates these dimensions using inclusion Bayes factors. The resulting model-averaged posterior estimates account for both parameter and model uncertainty, thereby improving the robustness of evidence synthesis. Recent software developments have made these methods accessible to applied researchers. JASP provides a user-friendly graphical interface, while R packages support reproducible workflows, automated reporting, and sensitivity analyses. This paper introduces the theoretical foundations and practical implementation of BMA and RoBMA, offering researchers a transparent and flexible framework for improving the credibility and replicability of meta-analytic conclusions.关键词
贝叶斯模型平均/元分析/模型不确定性/发表偏倚/稳健贝叶斯元分析Key words
bayesian model averaging/meta-analysis/model uncertainty/publication bias/robust bayesian meta-analysis引用本文复制引用
.基于贝叶斯模型平均的元分析：原理与实现[EB/OL].(2026-07-31)[2026-08-19].https://chinaxiv.org/abs/202605.00116.学科分类
自然科学研究方法
