Integrating Cross-Lagged Panel Models with Instrumental Variables to Extend the Temporal Generalizability of Causal Inference.
Integrating Cross-Lagged Panel Models with Instrumental Variables to Extend the Temporal Generalizability of Causal Inference.
复制标题
将交叉滞后面板模型与工具变量相集成，以扩展因果推理的时间普遍性。
DOI：
10.1080/00273171.2022.2160954
复制
发表时间：
2023
影响因子：
3.8
通讯作者：
Neale,MichaelC
中科院分区：
文献类型：
--
作者：
Singh,Madhurbain;Dolan,ConorV;Neale,MichaelC
交叉滞后面板模型（CLPM）中的滞后效应取决于重复测量之间的时间间隔（Kuiper & Ryan，2018），通常在较长的时间间隔内无法检测到。如果一项研究未能发现滞后效应，人们就无法区分这是由于缺乏因果效应还是时间间隔太长。因此，CLPM中缺乏因果关系的证据不能推广到两个变量之间的整体因果关系。为了解决这一问题，我们提出了一个模型集成工具变量（IV）在CLPM（以下简称IV-CLPM）。工具变量回归（IVR）利用假设暴露变量的外源预测因子（即IV）来估计其对结果的影响，而不需要对两个变量进行任何时间排序（Maydeu-Olivares等人，2019年）。因此，除了CLPM中传统估计的滞后（即“远端”）效应外，IV-CLPM（图1）允许基于IVR估计每个波处X和Y之间的横截面（即“近端”）效应。T1时的IVR估计近端效应（bXY 1和bYX 1）反映了首次评估前展开的因果过程。远端效应（bXY 12和bYX 12）表示给定T1和T2之间的时间间隔，X1对Y2和Y1对X2的影响。最后，T2时的近端效应（bXY 2和bYX 2）反映了研究期间累积但未被
The lagged effects in cross-lagged panel models (CLPM) depend on the time interval between repeated measures (Kuiper & Ryan, 2018), usually becoming undetectable at longer intervals. If a study fails to detect a lagged effect, one cannot distinguish whether it is due to the absence of causal effects or too long of a time interval. Therefore, a lack of evidence for causal influences in CLPM cannot be generalized to the overall causal relationship between two variables. To address this limitation, we present a model integrating instrumental variables (IVs) in CLPM (henceforth, IV-CLPM). Instrumental variables regression (IVR) utilizes exogenous predictors (ie, the IVs) of a hypothesized exposure variable to estimate its effect on the outcome, without needing any temporal ordering of the two variables (Maydeu-Olivares et al., 2019). Therefore, IV-CLPM (Figure 1) allows for IVR-based estimation of cross-sectional (ie,“proximal”) effects between X and Y at each wave, in addition to the lagged (ie,“distal”) effects traditionally estimated in CLPM. The IVR-estimated proximal effects at T1 (bXY1 and bYX1) reflect the causal process that unfolded before the first assessment. The distal effects (bXY12 and bYX12) represent the influence of X1 on Y2 and of Y1 on X2, given the time interval between T1 and T2. Lastly, the proximal effects at T2 (bXY2 and bYX2) reflect the causal influences that accumulated during the study period but were not captured by the