项目名称: 结构方程模型中基于充分降维技术的变量选择和模型诊断
项目编号: No.11261064
项目类型: 地区科学基金项目
立项/批准年度: 2013
项目学科: 数理科学和化学
项目作者: 陈飞
作者单位: 云南财经大学
项目金额: 45万元
中文摘要: 对于结构方程模型的统计推断是探索无法观测的潜在变量统计规律的重要工具,被广泛应用于行为学、心理学、生物学和医学等领域。在结构方程模型统计推断中,结构方程的设定是一个关键的步骤,包括对每个结构方程中所包含的自变量和方程形式的选择。目前,解决这个问题的主要手段包括模型比较,如:贝叶斯因子(Bayes Factor)、贝叶斯信息准则(BIC)、偏差信息准则(DIC),以及模型的拟合优度检验等。由于这些方法都需要待选模型的具体形式,因此,对于解决下面两个问题无法取得好的效果:(1)不依赖于结构方程具体形式,进行每个结构方程的解释变量的选择;(2)判断结构方程是否应为线性方程。充分降维理论和基于该理论的变量选择技术能够不依赖于模型具体形式而达到降维的 目的,但目前的方法需要变量的可观测样本,故本项目拟提出针对潜在变量的充分降维方法,以及基于该方法和压缩估计的结构方程变量选择技术,解决上述两个问题。
中文关键词: 结构方程模型;变量选择;模型诊断;充分降维;潜在变量
英文摘要: Statistical inferences for structural equation models are important approaches to reveal statistical information of latent variables which cannot be observed, and they are applied to various fields such as behavioural, psychological, biological, and medical sciences, and so on. In the structural equation modelling, the specification of structural equations is a crucial step, including the selection of explanatory variables and formula type in each structural equation. Till now, the main methodologies to solve this problem include model comparison such as Bayes factor, Bayesian Information Criterion (BIC), Deviance Information Criterion (DIC), and goodness-of-fit test of the model,and so on. As these methods depend on the specific formula of competing models, they cannot contribute much to solution of the following two problems: (1) explanatory variable selection in each structural equation without specification of its formula type; (2) whether the structural equation is linear or not. Theories of sufficient dimension reduction and approaches of variable selection based on them can solve the problem of dimension reduction without model specification, but currently, they are based on observed samples for variables. Hence, to solve the above two issues, this project will try to propose methodologies of sufficien
英文关键词: structural equation model;variable selection;model diagnostic;sufficient dimension reduction;latent variable