含潜变量模型下的因果效应研究

发布时间：2018-03-09 06:11

本文选题：潜变量　切入点：因果效应　出处：《长春工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：因果关系是复杂事物间普遍存在的一种基本关系,不仅在现实生活中存在着非常多的因果关系,在统计学、医学、经济学等众多领域中也得到了科研工作者们非常广泛的关注,有着重要的应用。在研究因果关系的过程中,潜变量的引入是非常重要的,它不仅更清楚并深刻地刻画了变量间的关系,同时为简化复杂问题做出了很大的贡献。对于含潜变量模型下因果效应的研究,本文主要从三个方面来进行。其一是虚拟事实模型,二是因果网模型,三就是基于因子分析进行因果关系的探索。第一部分是在虚拟事实模型框架下进行研究。我们在VanderWeele(2008)和Chiba(2009)对平均因果效应研究的基础上进行比较,通过实际例子和模拟研究,我们可以得出这种结论,在大多数情况下Chiba与VanderWeele的边界相比,前者的方法是更好的,因为他的方法可以得到一个较小的边界,这个小的边界可以使我们的研究更加精确,同时在实际应用中也可以得到更精确的结果。虚拟事实模型是通过引进虚拟事实变量,来评价一个变量对另一个变量的因果作用,而因果网模型是用网络结构表示变量间的因果关系,可以借助图模型的相关方法在潜变量存在的情况下判断变量间的因果关系(即网络中的边)是否存在。本篇文章通过NSCOT数据探索送往医院时间、身体状况、生存状况等变量之间的因果关系。我们可以得出送往医院的时间与生存状况是有因果关系的,而一个人的身体素质和受伤的严重情况会对病人送往医院的时间产生影响,从而对生存状况产生影响。第三个研究的问题是因子分析,因子分析的主要思想是用少数几个因子来描述多个指标之间的联系,以较少的几个因子反应原数据中的绝大部分信息的统计学方法,而这几个被综合的因子其实就是文章中讨论的潜变量。在本文中,利用各省市居民生活质量数据进行因子分析,经分析我们可以得到三个公共因子,其分别为总体因子、个体因子和消费品因子。这三个因子就是综合得来、无法直接观测到的潜变量。
[Abstract]:Causality is a kind of basic relationship that exists widely among complicated things. It not only has many causality in real life, but also has been widely concerned by researchers in many fields, such as statistics, medicine, economics and so on. In the process of studying causality, the introduction of latent variables is very important. It not only clearly and profoundly depicts the relationship between variables, At the same time, it has made a great contribution to simplify the complex problem. For the study of causality under the model with latent variables, this paper mainly deals with three aspects: one is the virtual fact model, the other is the causality net model. The first part is to carry on the research under the framework of the virtual fact model. We compare the average causality based on Vander Weele 2008) and Chibaan 2009), through the actual examples and simulation research. We can conclude that in most cases Chiba's method is better than VanderWeele's, because his method can get a smaller boundary, which makes our research more accurate. At the same time, more accurate results can be obtained in practical applications. Virtual fact models are used to evaluate the causal effect of one variable on another by introducing virtual fact variables. The causality net model uses the network structure to express the causality between variables. With the help of the relevant method of graph model, we can judge whether the causality between variables exists (that is, the edge of the network) in the presence of latent variables. This article explores the time to hospital, physical condition through NSCOT data. We can conclude that there is a causal relationship between the time taken to the hospital and the living condition, and that the physical quality of a person and the severity of the injury will have an impact on the time the patient is sent to the hospital. The third problem is factor analysis. The main idea of factor analysis is to use a few factors to describe the relationship between multiple indicators. A few factors reflect the vast majority of the information in the original data, and these combined factors are actually the underlying variables discussed in the paper. In this paper, factor analysis is carried out using the quality of life data of the residents of various provinces and cities. Through analysis, we can get three common factors, which are the total factor, the individual factor and the consumer factor, which are the latent variables which can not be observed directly.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：C81

【参考文献】