基于统计学角度:解读固定效应模型和随机效应模型

摘要

大多数的Meta分析都会用到固定效应模型和随机效应模型中的一种,固定效应模型假设所有的纳入研究拥有共同的真实效应量,而随机效应模型中的真实效应随纳入研究的不同而改变.运用的模型不同,所得到的合并后的效应量均数值也不相同,这不仅体现在效应量的均值上,更多的体现在每个纳入研究权重的分配上,本文主要目的是深度解剖两种模型以及两种模型的假设,区分其共同点和不同点,并通过两种模型计算每个研究所占的权重和合并后效应量的均数值,最后指出并比较其优缺点.%Most of Meta-analysis will use fixed effect model (FEM) or random effect model (REM), and FEM assumes that there is common true effect size in all included studies, while true effect size in REM will change according to different included studies. The mean of pooled effect size will be different as different models are used, which is not only reflected on the average, but also much on the weight distribution of each included study. Main purpose of this article is to deeply analyze these two models and their assumptions, distinguish their similarities and difference, calculate each study's weight and pooled effect size by these two models, and finally point out and compare their advantages and disadvantages.