On Subgrouping Continuous Processes in Discrete Time
On Subgrouping Continuous Processes in Discrete Time
复制标题
关于离散时间连续过程的子组
DOI：
10.1080/00273171.2022.2160957
复制
发表时间：
2023
影响因子：
3.8
通讯作者：
Molenaar, Peter C.
中科院分区：
文献类型：
--
作者：
Park, Jonathan J.;Fisher, Zachary;Chow, Sy-Miin;Molenaar, Peter C.
近年来，向量自回归（VAR）模型等动态网络模型及其各种形式在人类行为和心理研究中的应用蓬勃发展。VAR型模型在揭示各个领域的具体动态方面是非常宝贵的，例如影响（Wright等人，2019）和神经科学（Gates & Molenaar，2012）以及它们如何（错误）与规则发现一致。这些结果带来了一个新的理解的重要性N1研究理解个人动态。与这种具体应用的增长相对应的是，越来越多的人呼吁将个人特定领域与在群体水平上获得的推论相协调。已经开发了几种DT方法用于识别个体内动态中的异质亚组，包括诸如亚组组迭代多模型估计（S-GIMME; Gates & Molenaar，2012）程序和亚组链图形VAR（scGVAR; Park等人，但仍存在一些问题。基于VAR的方法具有一些局限性。例如，当时间间隔在研究中或研究之间变化时，可能会出现潜在的推理混淆。这些挑战可以通过在连续时间（CT）框架中拟合模型来改善;然而，由于大多数当代动态网络子分组方法中DT制剂的主要使用，文献中存在词汇缺口。这项工作探讨了各种基于DT的方法，如S-GIMME和scGVAR执行识别有意义的子组时，应用于连续过程。具体而言，我们评估了分组的准确性和质量的点估计，从这两种方法时，将它们应用到连续的过程中，在不同的采样间隔，DT，在蒙特卡洛模拟。我们解决了两个问题：（1）DT亚组方法在不同DT和效应量之间的准确性如何？不同的Dt s和效应量是否对应于点估计质量的可靠差异？我们模拟了来自4变量Ornstein-Uhlenbeck模型的CT数据，该模型结合了不同的设计因素，包括：采样间隔Dt，连续数据从Dt <$0.1 s到Dt <$10 s进行二次采样，数据生成CT模型的稳定程度，亚组之间的分离，以及时间样本量范围从T <$14到T <$100。
Recent years have witnessed a boom in the application of dynamic network models such as vector autoregression (VAR) models and their various flavors to the study of human behavior and psychology. VAR-type models have been invaluable in shedding light on idiographic dynamics in various areas such as affect (Wright et al., 2019) and neuroscience (Gates & Molenaar, 2012) and how they (mis-) align with nomothetic findings. These results have brought forth a new understanding of the importance of N ¼1 research for understanding individual dynamics. Commensurate to this rise in idiographic applications is an increased call for reconciling the person-specific domain with inferences obtained at the group-level. Several DT methodologies have been developed for identifying heterogeneous subgroups in intraindividual dynamics including approaches such as the subgroup group iterative multiple model estimation (S-GIMME; Gates & Molenaar, 2012) procedure and the subgrouped chain graphical VAR (scGVAR; Park et al., in press); however, some questions remain. VAR-based approaches bear some limitations. For instance, potential inferential confounds may arise when time intervals vary within or across studies. These challenges may be ameliorated by fitting models in the continuous-time (CT) framework; however, a lexical gap exists in the literature due to the predominant usage of DT formulations in most contemporary dynamic network subgrouping methods. This work examines how various DT-based approaches such as S-GIMME and the scGVAR perform at identifying meaningful subgroups when applied to continuous processes. Specifically, we assessed the subgrouping accuracy and the quality of point-estimates from these two methods when applying them onto continuous processes at various sampling intervals, Dt, in a Monte Carlo simulation. We addressed two questions:(1) How accurate are DT subgrouping approaches across different Dt s and effect sizes? and (2) Do different Dt s and effect sizes correspond to reliable differences in the quality of point estimates? We simulated CT data from 4-variate Ornstein–Uhlenbeck models incorporating varying design factors including: the sampling interval, Dt, that continuous data were subsampled from (Dt ¼ 0.1 s to Dt ¼ 10 sÞ, the degree of stability in the data-generating CT models, the separate between the subgroups, and the temporal sample size ranging from T ¼14 to T ¼100.
影响因子：
3.8
作者：
Park，Jonathan J.;Chow，Sy-Miin;Molenaar，Peter C. M.
通讯作者：
Molenaar，Peter C. M.