追踪研究中连续时间动态模型的建构、拓展与应用

在心理学研究中，基于离散时间的统计模型（如交叉滞后模型）常用于分析追踪设计下变量间的相互影响和因果关联。然而，该类模型假设不同次测量的时间间隔相等，且不同个体的测量时间点相同。实际中，这一假设很难满足，尤其在借助信息技术所获得的生态效度较高的复杂追踪设计中更是如此。本项目从心理学研究的需求和追踪设计所面临的问题出发，重点解决测量时间间隔不等、不同个体测量时间点相同或不同等复杂追踪设计下，用于探讨特质间相互影响及特质稳定性的连续时间动态模型的建构、拓展和应用的问题。研究内容主要包括：不同设计下连续时间动态模型的选择、建构及影响因素；连续时间动态模型中缺失数据的处理和测验结构稳定性的检验；含潜类别以及同时考虑特质发展和稳定性的连续时间动态模型的拓展；连续时间动态模型的软件实现和实际应用。本项目的研究将推动心理学研究更深入、合理地探索心理特质之间的动态发展过程及其因果关联。

Inferring reciprocal effect or causality between variables is a central goal of behavioral and psychological research. To address it, a variety of longitudinal dynamic models have been proposed. Most of these methods are discrete time (DT) models (e.g., cross-lagged panel models). DT models assume that time intervals between measurements are equal, and all subjects are assessed at the same set of time points. Violations of these assumptions are often ignored due to the difficulty of accounting for varying time intervals, which may lead to biased parameter estimates and ambiguous time course of effects. More importantly, parameters of DT models cannot be easily generalized or compared between studies when different time lags are used in these studies. However, the longitudinal data is often unequally spaced, especially for data collected by the experience sampling method or diary study, which poses a big challenge for DT modeling approaches. Continuous time (CT) dynamic models overcome the shortcomings of the DT models. First, they provide researchers flexibility in choosing data collection schemes. Second, they fit more naturally with theories and hypotheses about how processes are continually evolving. Finally, the parameters obtained by CT models can be compared or transformed into corresponding DT parameters for any given time intervals of interest. These advantages make the CT dynamic models a preferred approach, but this class of models is still in its infancy, and there remains plenty of problems to be answered in this field. Considering the actual needs and research questions of inferring reciprocal effects or causality between variables in psychological studies, this project aims to explore the construction, extension and application of the CT dynamic models based on the framework of the cross-lagged regression model. In detail, the main contents of this project are as follows: (1) Investigating the selection, construction, and influencing factors of the CT dynamic model under different longitudinal designs; (2) Developing suitable methods to handle missing data for the CT dynamic models, and then evaluating the validity of these methods to provide the researchers with suggestions on how to choose a proper method; (3) Extending the CT model to deal with more complex situations. Specifically, a CT dynamic model with measurement model, a mixed-CT dynamic model which can deal with latent classes, and an integration model of the CT dynamic model and the latent growth model which helps to understand the changes, stabilities, and reciprocal effects simultaneously will be developed; (5) Illustrating the application of the CT dynamic model using two real longitudinal datasets. This project will promote the development of the psychological research, and will help researchers to better understand the change of human behaviors and psychological traits as well as the causal relationships between variables.