密集型追踪数据中时变效应的分析及其在心理学研究中的应用

Intensive longitudinal data (ILD) are data with many measurements over time. As the development of smartphones and other technology, ILD are becoming more and more common. Due to the dense and comprehensive observations, ILD are expected to contain detailed information regarding temporal changes of behaviors, and allow researchers to explore change over time in the factors that influence an outcome. Meanwhile, the development of appropriate statistical methods to extract the information and capture the temporal changes of interest from ILD is needed. Time-varying effect model is widely used to analyze the intensive longitudinal data, for the reason that it is more flexible than other statistical model. However, due to the fact that this model is proposed recently, there has been little researches focusing on this method, leaving lots of important questions uninvestigated. . The project will focus on the statistical methods (especially the time-varying effect model) aimed at the analysis of ILD, the following issues will be studied: (1) The influence of “the number of repeated assessment” to the estimation accuracy of the time-varying effect model will be explored; What’s more, several factors such as the sample size, the type of growth pattern and the type of relationship between covariates and outcome will be extensively studied. (2) The model will be extended for categorical data and order data, and the necessity of model selection as well as the robustness of model estimation will be studied. (3) Both time-varying effect model and functional hierarchical linear model will be extended to including intervariability across individuals and latent classes, named mixed time-varying effects model, which lets researchers see changes in relationships between variables without making assumptions about the nature of those relationships, and explore the latent class as well as individual difference in growth pattern and (or) time-varying effects. (4) Different parameter estimations of extended will be compared using Monte Carlo simulation methods. For mixed time-varying effects model, Bayes estimation method will be provided and compared to other estimation methods. Meanwhile, the software will be developed. (5) Methods applications. This project will make ILD more appealing in the area of applied psychology, and help researchers to answer many interesting and significant questions that are previously unanswerable.

具有许多不同时间点的重复测量的数据称为密集型追踪数据。信息技术的发展使得这类数据的采集变得越来越容易，密集型追踪数据包含了关于行为随时间改变更详细的信息，不仅可以探索行为随时间变化的复杂模式，而且可以随时间变化的变量之间的影响效应，即时变效应。但是，由于数据结构的复杂性，在研究问题不断深入的同时其对数据分析方法也提出了新的挑战。本项目主要围绕用于密集型数据分析的时变效应模型，探讨如何应用此模型更好解决心理学应用研究的问题。包括（1）重复测量次数多少对时变效应模型参数估计精度的影响；（2）不同数据类型和分布形态下时变效应模型的拓展及其稳健性检验；（3）含有潜类别以及个体差异的统计分析模型的拓展；（4）参数估计方法的比较和程序实现；（5）实际案例分析。本项目的研究将为心理学的应用研究提供强有力的方法和易操作的工具，以使其更深入更合理地探索和理解人类行为的发展特点和规律。

行为科学研究的核心问题之一就是理解人类行为和心理过程在时间上的改变。追踪设计是探索特质发展、变化及个体差异常用的研究方法。密集型追踪数据对发展趋势和变量间关系的描述和解释更接近实际的情况。本项目围绕密集型追踪数据的特点，主要研究了以下内容：（1）分析密集型追踪数据的时变效应模型适用条件的问题。通过模拟研究的方法，系统探讨了重复测量的次数、样本量以及测量时间的分散程度对时变效应模型参数估计的影响。这一研究为拟采用时变效应模型的应用研究者，提供密集型追踪研究设计和数据采集方法方面的建议。（2）参数估计方法优劣及选择问题。结合时变效应模型，项目探讨了如何根据数据特点（如测量次数，时间间隔等），选择合适的估计方法（B样条和P样条）以及模型设定（如节点个数）的问题。项目还结合心理学研究中常用的模型，系统比较了不同模型下不同估计方法（如极大似然估计，限制性极大似然估计，贝叶斯估计）的优劣。研究结果为实际应用者提供如何结合不同的模型选取合适的参数估计方法的建议。（3）密集型追踪数据中缺失数据对时变效应模型参数估计精度的影响以及缺失值处理方法的选择。时变效应模型采用非参拟合的思路，对缺失数据采用简单的列删除方法。本项目探讨了列删除方法在不同缺失机制和不同缺失比例下的问题，创新性地提出了将参数假设下的基于模型的缺失数据处理方法用于时变效应模型。结果表明，采用参数建模的思路可分析密集型追踪数据的变效应模型，基于模型的缺失数据处理方法可用于处理时变效应模型的缺失数据。研究结果为应用者提供了新的时变效应模型建模方法和更合理的处理缺失数据的方法。（4）关于追踪研究设计中，项目还结合实际研究中可能遇到的问题进行了有针对性的拓展。如非连续发展特征和异质性群体的分析和探讨、基于动态测量过程性数据测量模型拓展等。（5）本项目提出的许多模型或方法在心理学实证研究中进行了应用，为实际应用者提供了参考和建议。