Detrending for Intensive Longitudinal Dyadic Data Analysis Using DSEM

Possible time trends are a common violation of the stationarity assumption, which is crucial in intensive longitudinal data analysis. Here we focus on the detrending issue in intensive longitudinal dyadic data analysis using Dynamic Structural Equation Modeling (DSEM). We first adjusted Savord et al. (2023) DSEM extension of the Actor-Partner Interdependence Model to better capture the interdependence between dyad members. Based on the adjusted model, using a simulation study, we investigated the influence of ignoring trends and compared two detrending practices—using residual DSEM (RDSEM) to separate time effects from the within-level autoregression or adding time covariate in autoregressive equations. Recommendations about whether and how to detrend are discussed.时变趋势是平稳性假设的常见违背情形，而平稳性假设是密集纵向数据分析的核心前提。本文聚焦于采用动态结构方程模型（Dynamic Structural Equation Modeling，DSEM）开展的密集纵向二元数据分析中的去趋势问题。本文首先对Savord等人（2023）提出的行动者-伴侣相互依赖模型（Actor-Partner Interdependence Model）的DSEM扩展形式进行了调整，以更精准地捕捉二元组成员间的相互依赖关系。基于调整后的模型，本文通过模拟实验探究了忽略趋势的影响，并对比了两种去趋势方法：一是采用残差动态结构方程模型（Residual DSEM，RDSEM）将时间效应与水平内自回归效应分离，二是在自回归方程中加入时间协变量。本文最后针对是否需要进行去趋势处理以及如何开展去趋势操作，给出了相应的建议。