Analyses of Overly Dispersed Covariance within Latent Structures and Applications in Psychological and Behavioral Research

本研究项目将研究多变量随时间的共变模型，并应用各种模型来分析社会科学和行为科学的数据。了解共变通常是研究导致特定结果的因果关系和可能机制的第一步。随着时间的推移，对共变的研究可能会揭示一个相互关联的变量系统中有趣而重要的模式。例如，在经济和金融领域，研究市场数据协变模式的变化对资产管理和投资组合多样化具有重要意义。世界各地的股市指数通常是相关的，但熊市和危机时期的相关性往往比正常时期高得多。认知领域的共变，如记忆、推理和信息处理速度，可用于评估早期认知障碍。例如，在年龄导致的认知能力下降的总体趋势中，不同领域的表现差异可能表明存在问题。该项目将为研究界开发工具，以促进对数据共变的解释。该项目还将培养研究生和博士后研究人员。本项目将探讨在潜在结构建模的背景下研究过度分散共变的不同方法。过度分散的共变是指驱动变量之间关联但未被规则潜在结构捕获的变异源。该项目将使用最先进的统计学和机器学习工具来检查来自社会和行为科学的数据。该项目的一个独特的智力贡献将是这些工具集适应社会和行为科学数据，这些数据通常强调多个结果变量，而不是像统计学和机器学习那样的多个预测变量。该项目将由几个示范应用组织，包括(1)态度调查中的反应一致性，(2)认知障碍模式，(3)儿童之间形成友谊的动态变化，以及(4)老年人的认知要求日常活动。这些应用在各自的内容上是广泛的。虽然它们说明了处理过度分散共变的不同和具体策略，但它们可以作为进一步开发其他研究领域类似应用的原型示例。

This research project will investigate models for covariation of multiple variables over time and apply the various models to analyze data from the social and behavioral sciences. Understanding covariation often is a first step in the study of causation and possible mechanisms that lead to specific outcomes. The study of covariation over time may reveal interesting and important patterns within a system of interconnecting variables. In economics and finance, for example, studying the change in patterns of covariation of market data has important implications for asset management and portfolio diversification. Stock-market indexes across the world often are correlated, but the correlations during bear markets and crisis periods tend to be much higher than during normal times. Covariation in cognitive domains, such as memory, reasoning, and speed of processing information may be used to assess early cognitive impairment. For example, divergence in performance across domains within the overall trend of general cognitive decline due to age could indicate problems. The project will develop tools for the research community to facilitate the interpretation of covariation in data. The project also will train graduate students and postdoctoral researchers.This project will examine different approaches for studying overly dispersed covariation in the context of modeling with latent structures. Overly dispersed covariation refers to sources of variation that drive the association between variables but are not captured by regular latent structures. The project will use state-of-the-art tools from statistics and machine learning to examine data from the social and behavioral sciences. A unique intellectual contribution of the project will be the adaptation of these toolsets to social and behavioral science data that often emphasize multiple outcome variables rather than multiple predictor variables as in the case of statistics and machine learning. The project will be organized by several exemplary applications including (1) response consistency in attitudinal survey, (2) patterns of cognitive impairment, (3) dynamics of change in forming friendship among children, and (4) cognitively demanding daily activities in older adults. The applications are broad in their respective content. While they illustrate different and specific strategies for handling overly dispersed covariation, they serve as prototypical examples for further development of similar applications in other fields of study.