Church C K, Schwenke J R
Control Clin Trials. 1986 Jun;7(2):149-64. doi: 10.1016/0197-2456(86)90030-9.
Repeated measures designs are common in clinical trials where recordings on a patient are made repeatedly over time. A basic assumption for the analysis of repeated measures designs is that a common correlation structure between observations for a given patient exists. When several observations are made for each patient this assumption may be tenuous at best. This article presents a pragmatic approach for combining a repeated measures design with a first-order autoregressive error component. A method of filtering the observed data to account for the autoregressive structure of the errors is considered. The effect on the analysis of variance results after extraction of the autoregressive component is seen by comparison of ANOVA summaries.
重复测量设计在临床试验中很常见，在这类试验中，会随着时间的推移对患者进行多次记录。重复测量设计分析的一个基本假设是，给定患者的观测值之间存在共同的相关结构。当对每个患者进行多次观测时，这个假设充其量可能是不成立的。本文提出了一种将重复测量设计与一阶自回归误差成分相结合的实用方法。考虑了一种对观测数据进行滤波以考虑误差自回归结构的方法。通过比较方差分析摘要，可以看出提取自回归成分后对方差分析结果的影响。
