贝叶斯混合效应模型:基于brms的应用教程

Bayesian Mixed-effects Models:A Primer with brms

相比传统方法,贝叶斯混合效应模型在处理数据层级结构和提供更直观的统计结果等方面有独特优势,这使其逐渐流行于心理学研究之中。然而,国内尚缺少将贝叶斯混合效应模型应用于心理学研究的介绍。本文先概述贝叶斯混合效应模型的基本概念和原理,然后结合模拟数据展示如何理解固定效应和随机效应,如何利用R语言的brms工具包设定和拟合贝叶斯混合效应模型,最后介绍如何借助先验预测检验自定义的先验分布是否合理,以及如何使用贝叶斯因子进行假设检验。凭借其强大的灵活性,贝叶斯混合效应模型可以适用于多样的心理学研究。

Compared to the traditional statistical methods,Bayesian linear mixed-effects modeling(BLMM)has a great number of advantages in dealing with the hierarchical structures underlying datasets and providing more intuitive statistical results.These advantages together popularize BLMM in psychological and other field research.However,there is still a lack of tutorials on the practical applications of BLMM in psychology studies in Chinese.Therefore,we first briefly introduced the basic concepts and rationales of BLMM.Then we employed a simulated dataset to demonstrate how to understand fixed effects and random effects,and how to use the popular brms R package to specify models for BLMM based on the experimental design.We additionally covered the procedure of pre-specifying priors with prior predictive checks,and the steps of performing hypothesis testing using the Bayes Factor.BLMM,with its extensions such as Generalized BLMM,has great flexibility and capability,they can and should be applied in various psychology research.